WO2020093630A1 - 一种基于多尺度深度语义分割网络的天线下倾角测量方法 - Google Patents

一种基于多尺度深度语义分割网络的天线下倾角测量方法 Download PDF

Info

Publication number
WO2020093630A1
WO2020093630A1 PCT/CN2019/076718 CN2019076718W WO2020093630A1 WO 2020093630 A1 WO2020093630 A1 WO 2020093630A1 CN 2019076718 W CN2019076718 W CN 2019076718W WO 2020093630 A1 WO2020093630 A1 WO 2020093630A1
Authority
WO
WIPO (PCT)
Prior art keywords
antenna
target
image
semantic segmentation
bounding box
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/076718
Other languages
English (en)
French (fr)
Inventor
翟懿奎
周集华
伍月婷
郑煜
徐颖
甘俊英
曾军英
邓文博
柯琪锐
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuyi University Fujian
Original Assignee
Wuyi University Fujian
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Wuyi University Fujian filed Critical Wuyi University Fujian
Priority to US16/652,346 priority Critical patent/US11561092B2/en
Priority to EP19856443.7A priority patent/EP3680609A4/en
Publication of WO2020093630A1 publication Critical patent/WO2020093630A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
    • G01B21/00—Measuring arrangements or details thereof, where the measuring technique is not covered by the other groups of this subclass, unspecified or not relevant
    • G01B21/22—Measuring arrangements or details thereof, where the measuring technique is not covered by the other groups of this subclass, unspecified or not relevant for measuring angles or tapers; for testing the alignment of axes
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/17—Terrestrial scenes taken from planes or by drones
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/176—Urban or other man-made structures
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Definitions

  • the invention relates to the field of mobile communication, and in particular to a method for measuring the downtilt angle of an antenna based on a multi-scale deep semantic segmentation network.
  • the quality of mobile communication networks is even more important.
  • the azimuth and downtilt of the antenna affect the coverage of the signal and the interference between the signals. Strict calculation and adjustment of the antenna need to be performed in time to improve the quality of the network signal.
  • the traditional methods for measuring the downtilt angle of the antenna can be divided into two types: one, manually climb to the antenna base station, and measure with a measuring instrument (compass, inclinometer, etc.); second, install an angle sensor on the antenna and return the data.
  • the antenna is susceptible to strong winds, heavy snow, and other factors, causing the downtilt angle to change, so it needs to be measured regularly.
  • the manual safety hazards and workload are large, and the practicability is low
  • the installation takes a long time and the antenna model is also not All are the same, so the cost of installation of the instrument is high, and the practicability is not high. Both of the above methods consume a lot of manpower and material resources and are not suitable for large-scale measurement today.
  • the object of the present invention is to provide an antenna downtilt measurement method based on a multi-scale deep semantic segmentation network, based on a drone as a carrier, calling a target detection algorithm and a semantic segmentation algorithm to measure the mobile base station antenna downtilt angle
  • the method has high applicability, low cost and safety.
  • An antenna downtilt measurement method based on multi-scale deep semantic segmentation network includes the following steps:
  • Image data collection the UAV is used to collect the base station antenna data, and the collected antenna image is used as the data set;
  • Target bounding box prediction locate the target antenna in the data set and use logistic regression to predict the bounding box;
  • Perform target recognition and semantic segmentation extract target features from the target antenna in the data set, learn the target features and perform activation function processing, output the target image for semantic image segmentation, and classify the pixel points of the target image and the background;
  • the image data collection includes:
  • the target bounding box prediction includes
  • Locate the target antenna in the antenna image use logistic regression to predict the bounding box, first divide the entire antenna image into N * N grids, predict the entire antenna image after the antenna image is input, and scan each one at a time Grid, when you locate the center of the grid where the target antenna is located, start predicting the target antenna.
  • Each bounding box predicts 4 coordinate values: t x , t y , t w , t h , the upper left of each target cell
  • the angular offset is (c x , c y ), and the heights of the bounding boxes are p x , p y respectively , then the network predicts the value as:
  • the input antenna image is divided into N * N grids, and each grid includes 5 predictors: (x, y, w, h, confidence) and a class c, so the network output is S * S * (5 * B + C) size; B is the number of bounding boxes in each grid, and C is the only antenna for the category of the present invention, so it is 1.
  • Confidence represents that the predicted grid contains the information about the confidence of the target antenna and the prediction accuracy of the bounding box:
  • target recognition and semantic segmentation include:
  • the antenna image is processed by the activation function after feature extraction. When the output value is greater than 0.5, the target is judged to be the antenna;
  • the output corresponding to each position i is y
  • the filter w, and the roundabout rate r are the steps of sampling the input signal.
  • a fully connected condition random field is used to classify the pixel points of the output target image, mainly to classify the boundary between the target image and the background.
  • the calculation of the antenna downtilt angle includes:
  • the downtilt angle of the base station antenna is the angle ⁇ between the base station antenna and the vertical plane
  • the beneficial effects of the present invention are: an antenna downtilt measurement method based on a multi-scale deep semantic segmentation network adopted by the present invention, based on a drone, using a target detection algorithm and a semantic segmentation algorithm to measure the mobile base station antenna downtilt angle
  • the method has high applicability, low cost and safety.
  • Figure 1 is a schematic diagram of the downtilt angle of the base station antenna
  • FIG. 2 is a flowchart of an antenna downtilt measurement method based on a multi-scale deep semantic segmentation network according to an embodiment of the present invention
  • FIG. 3 is a schematic diagram of frame prediction of an antenna downtilt measurement method based on a multi-scale deep semantic segmentation network according to an embodiment of the present invention
  • FIG. 4 is a schematic diagram of a network structure of an antenna downtilt measurement method based on a multi-scale deep semantic segmentation network according to an embodiment of the present invention
  • FIG. 5 is a schematic diagram of a bottleneck block of an antenna downtilt measurement method based on a multi-scale deep semantic segmentation network according to an embodiment of the present invention
  • FIG. 6 is a schematic diagram of standard convolution of an antenna downtilt measurement method based on a multi-scale deep semantic segmentation network according to an embodiment of the present invention
  • FIG. 7 is a schematic diagram of high-resolution feature extraction of an antenna downtilt measurement method based on a multi-scale deep semantic segmentation network according to an embodiment of the present invention
  • FIG. 8 is a schematic diagram of one-dimensional low-resolution feature extraction of an antenna downtilt measurement method based on a multi-scale deep semantic segmentation network according to an embodiment of the present invention
  • FIG. 9 is a schematic diagram of a hole convolution of a method for measuring the downtilt angle of an antenna based on a multi-scale deep semantic segmentation network according to an embodiment of the present invention.
  • FIG. 10 is a random field view of an antenna downtilt measurement method based on a multi-scale deep semantic segmentation network according to an embodiment of the present invention.
  • an embodiment of the present invention provides an antenna downtilt measurement method based on a multi-scale deep semantic segmentation network, including the following steps:
  • Image data collection the UAV is used to collect the base station antenna data, and the collected antenna image is used as the data set;
  • Target bounding box prediction locate the target antenna in the data set and use logistic regression to predict the bounding box;
  • Perform target recognition and semantic segmentation extract target features from the target antenna in the data set, learn the target features and perform activation function processing, output the target image for semantic image segmentation, and classify the pixel points of the target image and the background;
  • a method of measuring the downtilt angle of the mobile base station antenna by calling a target detection algorithm and a semantic segmentation algorithm has high applicability, low cost, and safety.
  • the image data collection includes:
  • the specific operations are: make the drone at the top of the pole of the base station antenna and record the latitude and longitude (L 0 , W 0 ) in the vertical direction of the pole; To fly, set its flight radius. The UAV moves around the pole along the radius on the same horizontal plane to obtain antenna images of different attitudes and angles of the mobile base station antenna as a data set.
  • the target bounding box prediction includes
  • Locate the target antenna in the antenna image use logistic regression to predict the bounding box, first divide the entire antenna image into N * N grids, predict the entire antenna image after the antenna image is input, and scan each one at a time Grid, when you locate the center of the grid where the target antenna is located, start predicting the target antenna.
  • Each bounding box predicts 4 coordinate values: t x , t y , t w , t h , the upper left of each target cell
  • the angular offset is (c x , c y )
  • the frame height of the bounding box is p x , p y respectively
  • the frame prediction is shown in Figure 3 then the network predicts the value for it:
  • the input antenna image is divided into N * N grids, and each grid includes 5 predictors: (x, y, w, h, confidence) and a class c, so the network output is S * S * (5 * B + C) size; B is the number of bounding boxes in each grid, and C is the only antenna for the category of the present invention, so it is 1.
  • Confidence represents that the predicted grid contains the information about the confidence of the target antenna and the prediction accuracy of the bounding box:
  • multi-scale prediction is used. There is no need to fix the size of the input image, so that different step sizes can be detected on feature maps of different sizes.
  • Three different detection layers are used to detect the antenna image on the target antenna, and different detection layers are realized by controlling the step size.
  • Use downsampling for the first detection layer using a step size of 32, which reduces the feature dimension. In order to connect to the previous same feature map, upsampling this layer can obtain a high resolution; the second use In the detection layer with a step size of 16, the remaining feature processing is the same as the first layer; in the third layer, the step size is set to 8, and feature prediction is performed on it. Finally, the detection accuracy of the target antenna is greater.
  • target recognition and semantic segmentation include:
  • the activation function is used in the logistic regression layer:
  • the antenna image is processed by the activation function after feature extraction. When the output value is greater than 0.5, the target is judged to be the antenna;
  • the network layer structure there are 0-74 layers, of which there are 53 convolutional layers and 22 residual layer networks.
  • 75-105 is the feature interaction layer of the neural convolution network, which can be divided into three scales, and the local feature interaction is realized by the convolution kernel.
  • the network structure is shown in FIG. 4.
  • the feature extraction of the perforated convolutional network is performed first; because the measured boundary accuracy is not high enough, the target image pixel must not be well separated from the background pixel, and it is combined with the fully connected conditional random field To improve the pixel classification of the image boundary, so that the segmentation effect is better.
  • the feature extraction of the network convolution layer can be divided into two cases: the input image with a low score is subjected to feature extraction using a standard convolution layer, see FIG. 6.
  • the high-resolution input image utilizes a detour convolution with a rate of 2 to perform dense feature extraction, refer to FIG. 7, and set its step size to 2, thereby reducing the feature dimension.
  • the convolution kernel is set to 3, the step size is 1, and the step size is 1.
  • FIG. 8 is a schematic diagram of one-dimensional low-resolution feature map extraction
  • FIG. 9 is a schematic diagram of hole convolution.
  • the output corresponding to each position i is y
  • the filter w, and the roundabout rate r are the steps of sampling the input signal, which can improve the receptive field of the filter.
  • the convolution with holes has an enlarged effect on the convolution kernel .
  • the residual module of multi-scale feature learning is used, and the bottleneck block is used in the present invention, and each convolution in the bottleneck block is processed by the normalization and activation function.
  • the bottleneck block is shown in Figure 5.
  • a fully connected condition random field is used to classify the pixel points of the output target image, mainly to classify the boundary between the target image and the background.
  • Each circle represents a pixel point
  • x i (white circle) is the pixel point (node) of the label
  • two-by-two is the edge of the pixel
  • y i black circle
  • x i the pixel point of the label
  • the classification of is determined by the reference value y i .
  • the image function output through the perforated convolutional network is a univariate potential function: Where the binary potential function is
  • the unary potential function function extracts the feature vector of a node of different feature graphs
  • the binary function connects the nodes extracted by the unary potential function, learns its edges, and connects all the nodes to form a fully connected layer conditional random field ,
  • the final output image of the function is more accurate.
  • the calculation of the antenna downtilt angle includes:
  • the downtilt angle of the base station antenna is the angle ⁇ between the base station antenna and the vertical plane

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Remote Sensing (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

本发明公开了一种基于多尺度深度语义分割网络的天线下倾角测量方法，采用无人机对基站天线数据进行采集，使用标注工具对获取到的天线图像进行标注，制作数据集；调用数据集进行训练并对模型进行调试；识别检测目标天线，将输出图像进行语义分割，最后获得最终分割的目标图像，再对目标图像下倾角进行计算，适用性高，成本低，安全。

Description

一种基于多尺度深度语义分割网络的天线下倾角测量方法 技术领域
本发明涉及移动通信领域，特别是一种基于多尺度深度语义分割网络的天线下倾角测量方法。
背景技术
如今为网络讯息时代，移动通信网络质量更是极为重要；在GSM-R建设及规划中，如图1所示，天线的方位角和下倾角影响着信号的覆盖范围以及信号之间的干扰，需要及时对天线进行严格的计算调整，进而提高网络信号的质量。
传统的测量天线下倾角方法可分为两种：一、人工攀爬至天线基站上，用测量仪(罗盘、坡度仪等)进行测量；二、在天线上安装角度传感器，回传数据。天线易受大风、大雪等其它因素的影响，导致下倾角改变，故需要定期的对其进行测量。而对于方法一，由于基站的高度较高，天线数量基数较大，其人工的安全隐患、工作量都较大，实用性较低；对于方法二，安装耗时较长，天线的型号也不尽相同，故在仪器的安装上成本较高，实用性不高。以上两种方法都消耗大量的人力物力，不适合如今大规模的测量。
发明内容
为解决上述问题，本发明的目的在于提供一种基于多尺度深度语义分割网络的天线下倾角测量方法，基于无人机为载体，调用目标检测算法和语义分割算法对移动基站天线下倾角进行测量的方法，适用性高，成本低，安全。
本发明解决其问题所采用的技术方案是：
一种基于多尺度深度语义分割网络的天线下倾角测量方法，包括以下步骤：
图像数据采集：采用无人机对基站天线数据进行采集，将并采集到的天线图像作为数据集；
目标边界框预测：对数据集中的目标天线进行定位，采用逻辑回归预测边界框；
进行目标识别和语义分割：对数据集中的目标天线进行目标特征提取，对目标特征进行学习并进行激活函数处理，输出目标图像进行语义图像分割，分类目标图像与背景的像素点；
计算天线下倾角：通过目标图像的边框得出天线框的宽度和高度，计算出天线下倾角。
进一步，所述图像数据采集，包括：
使无人机位于基站天线的抱杆顶端，记录抱杆竖直方向上的经纬度(L 0，W 0)；对基站天线进行绕点飞行，设置其飞行半径，无人机绕抱杆沿着半径在同一水平面上移动获取移动基站天线不同姿态和角度的天线图像作为数据集。
进一步，所述目标边界框预测，包括
对天线图像中的目标天线进行定位，使用逻辑回归预测边界框，先将整张天线图像划分成N*N个网格，在天线图像输入后对整张天线图像进行预测，一次性扫描每个网格，定位到目标天线所在网格中心时，开始对目标天线进行预测，每个边界框预测4个坐标值为：t x，t y，t w，t h，每个目标单元格的左上角偏移量为(c x，c y)，边界框的框高分别为p x，p y，则网络对其预测值为：
b x＝σ(t x)+c x                (1)
b y＝σ(t y)+c y               (2)
b w＝p we tw                (3)
b w＝p he tw                   (4)
输入的天线图像被分为N*N网格，每个网格包括5个预测量：(x，y，w，h，confidence)和一个c类，所以网络输出是S*S*(5*B+C)大小；B为每个网格中边界框数量，C对于本发明为类别只有天线，故为1。confidence代表了所预测的网格中含有目标天线的置信度和边界框的预测精度两个信息：
Figure PCTCN2019076718-appb-000001
设定阈值为0.5，当Pr(Object)＝1；目标天线落在格子中心，即当前预测的边界框与实际的背景框对象重合较之前更好；若预测边界框非当前最佳，阈值＜0.5时，便不对其进行预测边界框，判定目标天线没有落在网格中。
进一步，所述进行目标识别和语义分割，包括：
采用特征提取的网络卷积层进行目标识别：输入天线图像像素416*416，通道数为3，32层卷积核，每个核大小3*3，32层的卷积核，用于学习32种特征图，对于目标天线的颜色的差异利用不同的卷积核对目标天线特征进行学习；在特征提取时进行卷积层上采样，物体类别的预测公式如下：
Figure PCTCN2019076718-appb-000002
其中Pr(Class i|object)为物体类别可能性；
然后采用逻辑回归层运用激活函数：
Figure PCTCN2019076718-appb-000003
使得预测目标输出范围在0到1之间。天线图像经过特征提取后进过激活函数处理，当输出的值大于0.5时，便判断该目标为天线；
然后使用深度卷积网络对天线图像进行语义图像分割，分类目标图像与背景的像素点：
对目标图像进行输入后先经过带孔卷积网络的特征提取；输入特征图像后对空洞卷积进行计算：
y[i]＝∑ kx[i+r*k]*w[k]            (8)
对于二维信号，每个位置i对应输出为y，滤波器w，迂回速率r为输入信号进行采样的步长。
输入图像经过卷积网络处理输出后，采用全连接条件随机场对输出的目标图像的像素点进行分类处理，主要是对目标图像与背景边界的分类。
进一步，所述计算天线下倾角，包括：
通过目标图像的边框得出天线框的宽度x和高度y，运用几何关系对基站天线下倾角进行计算，基站天线下倾角为基站天线和垂直面的夹角θ，
Figure PCTCN2019076718-appb-000004
本发明的有益效果是：本发明采用的一种基于多尺度深度语义分割网络的天线下倾角测量方法，基于无人机为载体，调用目标检测算法和语义分割算法对移动基站天线下倾角进行测量的方法，适用性高，成本低，安全。
附图说明
下面结合附图和实例对本发明作进一步说明。
图1是基站天线下倾角的示意图；
图2是本发明一个实施例提供的一种基于多尺度深度语义分割网络的天线下倾角测量方法的流程图；
图3是本发明一个实施例提供的一种基于多尺度深度语义分割网络的天线下倾角测量方法的边框预测示意图；
图4是本发明一个实施例提供的一种基于多尺度深度语义分割网络的天线下倾角测量方法的网络结构示意图；
图5是本发明一个实施例提供的一种基于多尺度深度语义分割网络的天线下倾角测量方法的瓶颈块示意图；
图6是本发明一个实施例提供的一种基于多尺度深度语义分割网络的天线下倾角测量方法的标准卷积示意图；
图7是本发明一个实施例提供的一种基于多尺度深度语义分割网络的天线下倾角测量方法的高分辨率特征提取示意图；
图8是本发明一个实施例提供的一种基于多尺度深度语义分割网络的天线下倾角测量方法的一维低分辨率特征提取示意图；
图9是本发明一个实施例提供的一种基于多尺度深度语义分割网络的天线下倾角测量方法 的空洞卷积示意图；
图10是本发明一个实施例提供的一种基于多尺度深度语义分割网络的天线下倾角测量方法的随机场视图。
具体实施方式
参照图2，本发明的一个实施例提供了一种基于多尺度深度语义分割网络的天线下倾角测量方法，包括以下步骤：
图像数据采集：采用无人机对基站天线数据进行采集，将并采集到的天线图像作为数据集；
目标边界框预测：对数据集中的目标天线进行定位，采用逻辑回归预测边界框；
进行目标识别和语义分割：对数据集中的目标天线进行目标特征提取，对目标特征进行学习并进行激活函数处理，输出目标图像进行语义图像分割，分类目标图像与背景的像素点；
计算天线下倾角：通过目标图像的边框得出天线框的宽度和高度，计算出天线下倾角。
在本实施例中，基于无人机为载体，调用目标检测算法和语义分割算法对移动基站天线下倾角进行测量的方法，适用性高，成本低，安全。
进一步地，所述图像数据采集，包括：
使用无人机对基站天线数据进行采集，具体操作为：使无人机位于基站天线的抱杆顶端，记录抱杆竖直方向上的经纬度(L 0，W 0)；对基站天线进行绕点飞行，设置其飞行半径，无人机绕抱杆沿着半径在同一水平面上移动获取移动基站天线不同姿态和角度的天线图像作为数据集。
进一步地，所述目标边界框预测，包括
对天线图像中的目标天线进行定位，使用逻辑回归预测边界框，先将整张天线图像划分成N*N个网格，在天线图像输入后对整张天线图像进行预测，一次性扫描每个网格，定位到目标天线所在网格中心时，开始对目标天线进行预测，每个边界框预测4个坐标值为：t x，t y，t w，t h，每个目标单元格的左上角偏移量为(c x，c y)，边界框的框高分别为p x，p y，边框预测参照图3所示，则网络对其预测值为：
b x＝σ(t x)+c x                (1)
b y＝σ(t y)+c y               (2)
b w＝p we tw               (3)
b w＝p he tw                  (4)
输入的天线图像被分为N*N网格，每个网格包括5个预测量：(x，y，w，h，confidence)和一个c类，所以网络输出是S*S*(5*B+C)大小；B为每个网格中边界框数量，C对于本发明为类别只有天线，故为1。confidence代表了所预测的网格中含有目标天线的置信度和边界框的预测精度两个信息：
Figure PCTCN2019076718-appb-000005
设定阈值为0.5，当Pr(Object)＝1；目标天线落在格子中心，即当前预测的边界框与实际的背景框对象重合较之前更好；若预测边界框非当前最佳，阈值＜0.5时，便不对其进行预测边界框，判定目标天线没有落在网格中。
在目标的精准性上，运用多尺度预测。不需要固定输入图像的大小，从而可以在不同尺寸的特征图上用不同的步长进行检测。对目标天线使用三个不同检测层对天线图像进行检测，通过控制步长来实现不同的检测层。对第一个检测层采用下采样，使用步长为32，从而降低特征维度，为了与前一个相同特征图连接，再对该层进行上采样，此时可得到高的分辨率；第二使用步长为16的检测层，其余的特征处理与第一层一致；在第三层则将步长设置为8，对其进行特征预测，最后可得对目标天线的检测精确度更大。
进一步地，所述进行目标识别和语义分割，包括：
目标识别：
采用特征提取的网络卷积层进行目标识别：输入天线图像像素416*416，通道数为3，32层卷积核，每个核大小3*3，32层的卷积核，用于学习32种特征图，对于目标天线的颜色的差异利用不同的卷积核对目标天线特征进行学习；在特征提取时为了降低特征图的维度，进行卷积层上采样，物体类别的预测公式如下：
Figure PCTCN2019076718-appb-000006
其中Pr(Class i|object)为物体类别可能性；
然后，为了更加准确对目标天线进行识别，采用逻辑回归层运用激活函数：
Figure PCTCN2019076718-appb-000007
使得预测目标输出范围在0到1之间。天线图像经过特征提取后进过激活函数处理，当输出的值大于0.5时，便判断该目标为天线；
在网络层结构中，0-74层，其中有53个卷积层，22个残差层网络。75-105为神经卷积网络的特征交互层，其可分为三个尺度，以卷积核方式实现局部特征交互，其网络结构如图4所示。
在数据集的制作上，只对天线进行检测，因此类别为1。因此在训练时，最后一个卷积层的输出：3*(1+4+1)＝18。
语义分割：
使用深度卷积网络对天线图像进行语义图像分割，分类目标图像与背景的像素点：
对目标图像进行输入后先经过带孔卷积网络的特征提取；由于其测得边界精度不够高，使得目标图像像素不得与背景像素较好的分离，通过完全连接的条件随机场与之相结合来提高图像边界的像素分类，从而使得分割效果更佳。
先使用带孔卷积网络对其进行特征提取。对于网络卷积层对特征的提取可分为两种情况：低分率的输入图像使用标准卷积层对其进行特征提取，参照图6。高分辨率的输入图像利用速率为2的迂回卷积进行密集特征提取，参照图7，并且设置其步长为2，从而降低特征维度。在该卷积网络层设置其卷积核为3，步幅为1，步长为1。图8为一维低分辨率的特征图提取示意图，图9为空洞卷积示意图。
对于串行模块和空间金字塔池化层模块的网络结构中，带孔卷积可以有效增加滤波器的感受野，整合多尺度信息。输入特征图像后对空洞卷积进行计算：
y[i]＝∑ kx[i+r*k]*w[k]            (8)
对于二维信号，每个位置i对应输出为y，滤波器w，迂回速率r为输入信号进行采样的步长，可以提高滤波器的感受野，带孔卷积对于卷积核有着扩大的效果。在特征网络的提取上，使用了多尺度特征学习的残差模块，而在本发明中运用的为瓶颈块，在瓶颈块中每个卷积都经过归一化和激活函数的处理。从而使得上下文的语境信息更加丰富，瓶颈块如图5。
输入图像经过卷积网络处理输出后，采用全连接条件随机场对输出的目标图像的像素点进行分类处理，主要是对目标图像与背景边界的分类。
随机场视图如图10。每个圆圈代表像素点，x i(白色圆圈)为受标签的像素点(节点)，两两相连的为像素的边，y i(黑色圆圈)为x i的参考值，所标签的像素点的分类则通过参考值y i来判定。由吉布斯分布函数：
Figure PCTCN2019076718-appb-000008
其中y为像素点x i的参考值，E(y|I)为能量函数。
Figure PCTCN2019076718-appb-000009
通过带孔卷积网络输出的图像函数为一元势函数：
Figure PCTCN2019076718-appb-000010
其中二元势函数为
Figure PCTCN2019076718-appb-000011
该函数处理像素点之间的关系，将像同样的素点赋予相同的符号。一元势函数函数提取不同特征图的一个节点的特征向量，二元函数便将一元势函数所提取的节点相连接，对其边进行学习，将所有的节点连接起来便为全连接层条件随机场，函数最终输出的图像更加精准。
进一步，所述计算天线下倾角，包括：
通过目标图像的边框得出天线框的宽度x和高度y，运用几何关系对基站天线下倾角进行计算，基站天线下倾角为基站天线和垂直面的夹角θ，
Figure PCTCN2019076718-appb-000012
以上所述，只是本发明的较佳实施例而已，本发明并不局限于上述实施方式，只要其以相同的手段达到本发明的技术效果，都应属于本发明的保护范围。

Claims (5)

  1. 一种基于多尺度深度语义分割网络的天线下倾角测量方法，其特征在于，包括以下步骤：
    图像数据采集：采用无人机对基站天线数据进行采集，将并采集到的天线图像作为数据集；
    目标边界框预测：对数据集中的目标天线进行定位，采用逻辑回归预测边界框；
    进行目标识别和语义分割：对数据集中的目标天线进行目标特征提取，对目标特征进行学习并进行激活函数处理，输出目标图像进行语义图像分割，分类目标图像与背景的像素点；
    计算天线下倾角：通过目标图像的边框得出天线框的宽度和高度，计算出天线下倾角。
  2. 根据权利要求1所述的一种基于多尺度深度语义分割网络的天线下倾角测量方法，其特征在于，所述图像数据采集，包括：
    使无人机位于基站天线的抱杆顶端，记录抱杆竖直方向上的经纬度(L 0，W 0)；对基站天线进行绕点飞行，设置其飞行半径，无人机绕抱杆沿着半径在同一水平面上移动获取移动基站天线不同姿态和角度的天线图像作为数据集。
  3. 根据权利要求2所述的一种基于多尺度深度语义分割网络的天线下倾角测量方法，其特征在于，所述目标边界框预测，包括
    对天线图像中的目标天线进行定位，使用逻辑回归预测边界框，先将整张天线图像划分成N*N个网格，在天线图像输入后对整张天线图像进行预测，一次性扫描每个网格，定位到目标天线所在网格中心时，开始对目标天线进行预测，每个边界框预测4个坐标值为：t x，t y，t w，t h，每个目标单元格的左上角偏移量为(c x，c y)，边界框的框高分别为p x，p y，则网络对其预测值为：
    b x＝σ(t x)+c x  (1)
    b y＝σ(t y)+c y  (2)
    b w＝p we tw (3)
    b w＝p he tw   (4)
    输入的天线图像被分为N*N网格，每个网格包括5个预测量：(x，y，w，h，confidence)和一个c类，所以网络输出是S*S*(5*B+C)大小；B为每个网格中边界框数量，C对于本发明为类别只有天线，故为1。confidence代表了所预测的网格中含有目标天线的置信度和边界框的预测精度两个信息：
    Figure PCTCN2019076718-appb-100001
    设定阈值为0.5，当Pr(Object)＝1；目标天线落在格子中心，即当前预测的边界框与实际的背景框对象重合较之前更好；若预测边界框非当前最佳，阈值＜0.5时，便不对其进行预测边界框，判定目标天线没有落在网格中。
  4. 根据权利要求3所述的一种基于多尺度深度语义分割网络的天线下倾角测量方法，其特征在于，所述进行目标识别和语义分割，包括：
    采用特征提取的网络卷积层进行目标识别：输入天线图像像素416*416，通道数为3，32层卷积核，每个核大小3*3，32层的卷积核，用于学习32种特征图，对于目标天线的颜色的差异利用不同的卷积核对目标天线特征进行学习；在特征提取时进行卷积层上采样，物体类别 的预测公式如下：
    Figure PCTCN2019076718-appb-100002
    其中Pr(Class i|object)为物体类别可能性；
    然后采用逻辑回归层运用激活函数：
    Figure PCTCN2019076718-appb-100003
    使得预测目标输出范围在0到1之间。天线图像经过特征提取后进过激活函数处理，当输出的值大于0.5时，便判断该目标为天线；
    然后使用深度卷积网络对天线图像进行语义图像分割，分类目标图像与背景的像素点：
    对目标图像进行输入后先经过带孔卷积网络的特征提取；输入特征图像后对空洞卷积进行计算：
    Figure PCTCN2019076718-appb-100004
    对于二维信号，每个位置i对应输出为y，滤波器w，迂回速率r为输入信号进行采样的步长。
    输入图像经过卷积网络处理输出后，采用全连接条件随机场对输出的目标图像的像素点进行分类处理，主要是对目标图像与背景边界的分类。
  5. 根据权利要求4所述的一种基于多尺度深度语义分割网络的天线下倾角测量方法，其特征在于，所述计算天线下倾角，包括：
    通过目标图像的边框得出天线框的宽度x和高度y，运用几何关系对基站天线下倾角进行计算，基站天线下倾角为基站天线和垂直面的夹角θ，
    Figure PCTCN2019076718-appb-100005
PCT/CN2019/076718 2018-11-09 2019-03-01 一种基于多尺度深度语义分割网络的天线下倾角测量方法 Ceased WO2020093630A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US16/652,346 US11561092B2 (en) 2018-11-09 2019-03-01 Method for measuring antenna downtilt angle based on multi-scale deep semantic segmentation network
EP19856443.7A EP3680609A4 (en) 2018-11-09 2019-03-01 METHOD OF MEASURING THE DOWNWARD ANGLE OF AN ANTENNA BASED ON A MULTI-SCALE DEEP SEMANTIC SEGMENTATION NETWORK

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811338415.4 2018-11-09
CN201811338415.4A CN109685762A (zh) 2018-11-09 2018-11-09 一种基于多尺度深度语义分割网络的天线下倾角测量方法

Publications (1)

Publication Number Publication Date
WO2020093630A1 true WO2020093630A1 (zh) 2020-05-14

Family

ID=66184673

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/076718 Ceased WO2020093630A1 (zh) 2018-11-09 2019-03-01 一种基于多尺度深度语义分割网络的天线下倾角测量方法

Country Status (4)

Country Link
US (1) US11561092B2 (zh)
EP (1) EP3680609A4 (zh)
CN (1) CN109685762A (zh)
WO (1) WO2020093630A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112132965A (zh) * 2020-09-25 2020-12-25 中国矿业大学 一种岩土体孔裂隙结构多尺度表征方法
CN112329808A (zh) * 2020-09-25 2021-02-05 武汉光谷信息技术股份有限公司 一种Deeplab语义分割算法的优化方法及系统
CN112784857A (zh) * 2021-01-29 2021-05-11 北京三快在线科技有限公司 一种模型训练以及图像处理方法及装置
CN113239815A (zh) * 2021-05-17 2021-08-10 广东工业大学 一种基于真实语义全网络学习的遥感影像分类方法、装置及设备
CN113450311A (zh) * 2021-06-01 2021-09-28 国网河南省电力公司漯河供电公司 基于语义分割和空间关系的带销螺丝缺陷检测方法及系统
CN115170667A (zh) * 2022-07-15 2022-10-11 浙江大学 一种基于深度学习的无水尺塞氏盘水质透明度检测方法
CN116051843A (zh) * 2023-02-07 2023-05-02 智洋创新科技股份有限公司 一种熔融流股倾斜定量分析方法和系统
CN120121470A (zh) * 2025-05-15 2025-06-10 中南大学 一种适用于非垂直拍摄的浆液流动度图像的识别方法及装置

Families Citing this family (33)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109579774B (zh) * 2018-11-06 2021-04-23 五邑大学 一种基于深度实例分割网络的天线下倾角测量方法
WO2020194378A1 (ja) * 2019-03-22 2020-10-01 日本電気株式会社 画像処理システム、画像処理装置、画像処理方法、及びコンピュータ可読媒体
CN110647794B (zh) * 2019-07-12 2023-01-03 五邑大学 基于注意力机制的多尺度sar图像识别方法及装置
CN110688904A (zh) * 2019-08-30 2020-01-14 中通服建设有限公司 基于5g无人机的基站天线工参勘测方法及装置
CN110930417B (zh) * 2019-11-26 2023-08-08 腾讯科技（深圳）有限公司 图像分割模型的训练方法和装置、图像分割方法和装置
EP4070238B1 (en) * 2020-01-08 2024-09-11 Google LLC Translation of text depicted in images
CN111340139B (zh) * 2020-03-27 2024-03-05 中国科学院微电子研究所 一种图像内容复杂度的判别方法及装置
CN111539983B (zh) * 2020-04-15 2023-10-20 上海交通大学 基于深度图像的运动物体分割方法及系统
CN111768392B (zh) * 2020-06-30 2022-10-14 创新奇智(广州)科技有限公司 目标检测方法及装置、电子设备、存储介质
CN111860297A (zh) * 2020-07-17 2020-10-30 厦门理工学院 一种应用于室内固定空间的slam回环检测方法
CN112171668A (zh) * 2020-09-21 2021-01-05 河南颂达信息技术有限公司 一种基于人工智能的轨道式机器人防卡死检测方法及装置
CN112965369B (zh) * 2021-01-29 2022-02-01 电子科技大学 无人机三维无线通信网络的分布式部署方法
CN113538476B (zh) * 2021-07-20 2024-05-28 大连民族大学 一种基于边缘特征提取的深度学习图像分割方法及系统
CN114413855B (zh) * 2022-01-25 2023-04-07 湖南大学 一种基于图像识别的移动基站天线下倾角检测方法和系统
US12586198B2 (en) * 2022-02-18 2026-03-24 Techcyte, Inc. Image analysis for identifying objects and classifying background exclusions
CN114549958B (zh) * 2022-02-24 2023-08-04 四川大学 基于上下文信息感知机理的夜间和伪装目标检测方法
CN116758147B (zh) * 2022-03-02 2025-11-11 上海交通大学 基于手机信息融合的板状天线下倾角测量方法及系统
CN114322793B (zh) * 2022-03-16 2022-07-15 科大天工智能装备技术（天津）有限公司 基于全局分割网络的工件尺寸测量方法、装置及存储介质
CN114638991A (zh) * 2022-03-18 2022-06-17 聊城大学 一种基于YOLOv4-MobileNetv3和PyQt5的智能零售结算方法
CN114677510B (zh) * 2022-03-22 2025-04-15 中南大学 特征图上采样方法、小目标语义分割方法及成像方法
US12159465B2 (en) * 2022-04-14 2024-12-03 Toyota Research Institute, Inc. End-to-end learned lane boundary detection based on a transformer
CN114743010B (zh) * 2022-06-13 2022-08-26 山东科技大学 基于深度学习的超高压输电线路点云数据语义分割方法
CN115187959B (zh) * 2022-07-14 2023-04-14 清华大学 一种基于双目视觉的飞行汽车山地着陆方法及系统
CN114993243A (zh) * 2022-08-04 2022-09-02 深圳粤讯通信科技有限公司 一种基于物联网的天线姿态监控预警系统
CN115393725B (zh) * 2022-10-26 2023-03-07 西南科技大学 一种特征增强和语义分割的桥梁裂缝识别方法
CN116052003B (zh) * 2023-02-07 2024-05-14 中科星图数字地球合肥有限公司 一种天线角度信息的测量方法、装置及相关设备
CN116258817B (zh) * 2023-02-16 2024-01-30 浙江大学 一种基于多视图三维重建的自动驾驶数字孪生场景构建方法和系统
CN116612282B (zh) * 2023-05-26 2025-09-05 东风悦享科技有限公司 一种新的无人公交系统快速语义分割方法、系统及存储介质
CN117078760B (zh) * 2023-09-18 2025-11-11 北方民族大学 基于图像处理的阀体重心定位方法
CN117994797B (zh) * 2024-04-02 2024-06-21 杭州海康威视数字技术股份有限公司 一种水尺读数方法、装置、存储介质和电子设备
CN118823551A (zh) * 2024-09-19 2024-10-22 干将未来科技产业(杭州)有限公司 一种基于深度学习技术的ai算法优化系统及方法
CN119741293B (zh) * 2024-12-26 2025-09-30 中国舰船研究设计中心 一种基于图像识别的船舶腐蚀快速评估方法
CN120947588A (zh) * 2025-10-15 2025-11-14 浙江点创信息科技有限公司 基站天线姿态的核查方法、装置、计算机设备及存储介质

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103256920A (zh) * 2012-02-15 2013-08-21 天宝导航有限公司 利用图像处理测定倾斜角和倾斜方向
CN103630107A (zh) * 2012-08-23 2014-03-12 北京交通大学 一种基站天线倾角测量方法及数据处理方法
US20140205205A1 (en) * 2011-07-01 2014-07-24 Thomas Neubauer Method and apparatus for determining and storing the position and orientation of antenna structures
CN104504381A (zh) * 2015-01-09 2015-04-08 博康智能网络科技股份有限公司 非刚体目标检测方法及其系统
US9596617B2 (en) * 2015-04-14 2017-03-14 ETAK Systems, LLC Unmanned aerial vehicle-based systems and methods associated with cell sites and cell towers
CN106683091A (zh) * 2017-01-06 2017-05-17 北京理工大学 一种基于深度卷积神经网络的目标分类及姿态检测方法
CN107664491A (zh) * 2016-07-28 2018-02-06 中国电信股份有限公司 基站天线下倾角测量方法、装置和系统
US9918235B2 (en) * 2015-11-24 2018-03-13 Verizon Patent And Licensing Inc. Adaptive antenna operation for UAVs using terrestrial cellular networks
CN107830846A (zh) * 2017-09-30 2018-03-23 杭州艾航科技有限公司 一种利用无人机和卷积神经网络测量通信塔天线角度方法

Family Cites Families (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
PE20030995A1 (es) * 2002-01-24 2003-11-29 Telecom Italia Mobile Spa Metodo de registro de coordenadas geograficas, azimut e inclinacion de antena en una estacion de radio base para telefonia movil
US8374979B2 (en) * 2009-11-18 2013-02-12 Nec Laboratories America, Inc. Fast image parsing by graph adaptive dynamic programming (GADP) performing classification, detection, and segmentation simultaneously
US20110150317A1 (en) * 2009-12-17 2011-06-23 Electronics And Telecommunications Research Institute System and method for automatically measuring antenna characteristics
US9842274B2 (en) * 2014-03-28 2017-12-12 Xerox Corporation Extending data-driven detection to the prediction of object part locations
JP6443700B2 (ja) * 2014-05-27 2018-12-26 華為技術有限公司Ｈｕａｗｅｉ Ｔｅｃｈｎｏｌｏｇｉｅｓ Ｃｏ．，Ｌｔｄ． アンテナ設定パラメータを取得する方法および装置、ならびにシステム
US9855658B2 (en) * 2015-03-19 2018-01-02 Rahul Babu Drone assisted adaptive robot control
CN104978580B (zh) * 2015-06-15 2018-05-04 国网山东省电力公司电力科学研究院 一种用于无人机巡检输电线路的绝缘子识别方法
CN106851665A (zh) * 2015-12-07 2017-06-13 上海无线通信研究中心 天线的下倾角调整方法和基站
US10657364B2 (en) * 2016-09-23 2020-05-19 Samsung Electronics Co., Ltd System and method for deep network fusion for fast and robust object detection
CN115097937B (zh) * 2016-11-15 2025-04-29 奇跃公司 用于长方体检测的深度学习系统
CN106709568B (zh) * 2016-12-16 2019-03-22 北京工业大学 基于深层卷积网络的rgb-d图像的物体检测和语义分割方法
US10565787B1 (en) * 2017-01-27 2020-02-18 NHIAE Group, LLC Systems and methods for enhanced 3D modeling of a complex object
WO2018144650A1 (en) * 2017-01-31 2018-08-09 Focal Systems, Inc. Automated checkout system through mobile shopping units
US10678846B2 (en) * 2017-03-10 2020-06-09 Xerox Corporation Instance-level image retrieval with a region proposal network
US10402689B1 (en) * 2017-04-04 2019-09-03 Snap Inc. Generating an image mask using machine learning
US11257198B1 (en) * 2017-04-28 2022-02-22 Digimarc Corporation Detection of encoded signals and icons
EP3432263B1 (en) * 2017-07-17 2020-09-16 Siemens Healthcare GmbH Semantic segmentation for cancer detection in digital breast tomosynthesis
US10474988B2 (en) * 2017-08-07 2019-11-12 Standard Cognition, Corp. Predicting inventory events using foreground/background processing
WO2019066794A1 (en) * 2017-09-27 2019-04-04 Google Llc END-TO-END NETWORK MODEL FOR HIGH-RESOLUTION IMAGE SEGMENTATION
US10872228B1 (en) * 2017-09-27 2020-12-22 Apple Inc. Three-dimensional object detection
US20190130189A1 (en) * 2017-10-30 2019-05-02 Qualcomm Incorporated Suppressing duplicated bounding boxes from object detection in a video analytics system
US10878294B2 (en) * 2018-01-05 2020-12-29 Irobot Corporation Mobile cleaning robot artificial intelligence for situational awareness
US10649459B2 (en) * 2018-04-26 2020-05-12 Zoox, Inc. Data segmentation using masks
US10720058B2 (en) * 2018-09-13 2020-07-21 Volvo Car Corporation System and method for camera or sensor-based parking spot detection and identification
US11948362B2 (en) * 2018-10-31 2024-04-02 Arcus Holding A/S Object detection using a combination of deep learning and non-deep learning techniques

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140205205A1 (en) * 2011-07-01 2014-07-24 Thomas Neubauer Method and apparatus for determining and storing the position and orientation of antenna structures
CN103256920A (zh) * 2012-02-15 2013-08-21 天宝导航有限公司 利用图像处理测定倾斜角和倾斜方向
CN103630107A (zh) * 2012-08-23 2014-03-12 北京交通大学 一种基站天线倾角测量方法及数据处理方法
CN104504381A (zh) * 2015-01-09 2015-04-08 博康智能网络科技股份有限公司 非刚体目标检测方法及其系统
US9596617B2 (en) * 2015-04-14 2017-03-14 ETAK Systems, LLC Unmanned aerial vehicle-based systems and methods associated with cell sites and cell towers
US9918235B2 (en) * 2015-11-24 2018-03-13 Verizon Patent And Licensing Inc. Adaptive antenna operation for UAVs using terrestrial cellular networks
CN107664491A (zh) * 2016-07-28 2018-02-06 中国电信股份有限公司 基站天线下倾角测量方法、装置和系统
CN106683091A (zh) * 2017-01-06 2017-05-17 北京理工大学 一种基于深度卷积神经网络的目标分类及姿态检测方法
CN107830846A (zh) * 2017-09-30 2018-03-23 杭州艾航科技有限公司 一种利用无人机和卷积神经网络测量通信塔天线角度方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3680609A4 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112132965A (zh) * 2020-09-25 2020-12-25 中国矿业大学 一种岩土体孔裂隙结构多尺度表征方法
CN112329808A (zh) * 2020-09-25 2021-02-05 武汉光谷信息技术股份有限公司 一种Deeplab语义分割算法的优化方法及系统
CN112132965B (zh) * 2020-09-25 2024-03-26 中国矿业大学 一种岩土体孔裂隙结构多尺度表征方法
CN112784857A (zh) * 2021-01-29 2021-05-11 北京三快在线科技有限公司 一种模型训练以及图像处理方法及装置
CN113239815A (zh) * 2021-05-17 2021-08-10 广东工业大学 一种基于真实语义全网络学习的遥感影像分类方法、装置及设备
CN113450311A (zh) * 2021-06-01 2021-09-28 国网河南省电力公司漯河供电公司 基于语义分割和空间关系的带销螺丝缺陷检测方法及系统
CN115170667A (zh) * 2022-07-15 2022-10-11 浙江大学 一种基于深度学习的无水尺塞氏盘水质透明度检测方法
CN116051843A (zh) * 2023-02-07 2023-05-02 智洋创新科技股份有限公司 一种熔融流股倾斜定量分析方法和系统
CN120121470A (zh) * 2025-05-15 2025-06-10 中南大学 一种适用于非垂直拍摄的浆液流动度图像的识别方法及装置

Also Published As

Publication number Publication date
EP3680609A4 (en) 2021-02-24
EP3680609A1 (en) 2020-07-15
US20210215481A1 (en) 2021-07-15
US11561092B2 (en) 2023-01-24
CN109685762A (zh) 2019-04-26

Similar Documents

Publication Publication Date Title
WO2020093630A1 (zh) 一种基于多尺度深度语义分割网络的天线下倾角测量方法
CN106127204B (zh) 一种全卷积神经网络的多方向水表读数区域检测算法
CN110703800A (zh) 基于无人机的电力设施智能识别方法及系统
KR102507501B1 (ko) 인공지능 기반 수계오염원 모니터링 시스템 및 방법
CN109255317B (zh) 一种基于双网络的航拍图像差异检测方法
CN113325868B (zh) 一种基于无人机的农作物实时识别系统及方法
CN108052940A (zh) 基于深度学习的sar遥感图像水面目标检测方法
CN108711172B (zh) 基于细粒度分类的无人机识别与定位方法
CN114252868A (zh) 激光雷达的标定方法、装置、计算机设备和存储介质
CN111553280A (zh) 基于深度学习的目标部位识别方法
CN114252883A (zh) 目标检测方法、装置、计算机设备和介质
CN114252884A (zh) 路侧雷达定位监测方法、装置、计算机设备和存储介质
WO2020093624A1 (zh) 一种基于多尺度检测算法的天线下倾角测量方法
CN108932273A (zh) 图片筛选方法及装置
CN120526199B (zh) 基于无人机与机器视觉技术的边坡病害识别的方法和装置
Chen et al. BARS: a benchmark for airport runway segmentation
CN120403640A (zh) 一种基于深度学习的无人机测绘方法及系统
CN111414855A (zh) 基于端到端回归模型的电线杆标牌目标检测及识别方法
CN116258956A (zh) 无人机树木识别方法、设备、存储介质及装置
CN118298326B (zh) 基于深度学习和无人机遥感的海洋垃圾面积估算方法
CN118587600A (zh) 基于改进的Mask R-CNN机场跑道遥感影像目标检测分割方法
CN114252859A (zh) 目标区域的确定方法、装置、计算机设备和存储介质
CN109829923B (zh) 一种基于深度神经网络的基站天线下倾角测量系统及方法
CN116597315A (zh) 基于YOLO v3-tiny的红外遥感图像跑道检测方法
CN114255264A (zh) 多基站配准方法、装置、计算机设备和存储介质

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2019856443

Country of ref document: EP

Effective date: 20200408

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19856443

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE