CN-122023864-A - Roof photovoltaic anomaly identification method and system based on multidimensional proprietary features

CN122023864ACN 122023864 ACN122023864 ACN 122023864ACN-122023864-A

Abstract

The invention relates to a roof photovoltaic abnormality identification method and system based on multi-dimensional exclusive characteristics, wherein the method comprises the steps of obtaining a roof photovoltaic panel image and preprocessing the roof photovoltaic panel image; the method comprises the steps of carrying out multi-dimensional feature extraction on a preprocessed roof photovoltaic panel image to obtain multi-dimensional feature vectors, wherein the multi-dimensional features comprise bird dung shielding candidate area masks, cleaning panel exclusive features, bird dung shielding exclusive features, dust pollution exclusive features, panel cracking exclusive features and general auxiliary features, and carrying out roof photovoltaic anomaly identification according to the multi-dimensional feature vectors by adopting a machine learning model. Compared with the prior art, the method and the device realize high-precision identification of four types of abnormalities, namely cleaning of the plate body, shielding of bird droppings, dust pollution and plate body cracks, and simultaneously meet the low resource consumption requirement of the roof photovoltaic edge monitoring terminal.

Inventors

JIANG BENJIAN
HAN DONG
LIU XIN
WANG YINCHAO
YU JUNXIA
GU WEN
SHEN JINGJING
YUE YINGBO
CUI HAOYANG

Assignees

国网上海市电力公司

Dates

Publication Date: 20260512
Application Date: 20251209

Claims (10)

1. The roof photovoltaic abnormality identification method based on the multidimensional proprietary features is characterized by comprising the following steps of: acquiring a roof photovoltaic panel image, and preprocessing the roof photovoltaic panel image; Carrying out multi-dimensional feature extraction on the preprocessed roof photovoltaic panel image to obtain multi-dimensional feature vectors, wherein the multi-dimensional features comprise cleaning panel body exclusive features, bird dung shielding exclusive features, dust pollution exclusive features, panel body cracking exclusive features and general auxiliary features; and carrying out roof photovoltaic anomaly identification according to the multidimensional feature vector by adopting a machine learning model.
2. The method for identifying the photovoltaic anomalies on the roof based on the multi-dimensional exclusive features according to claim 1, wherein the exclusive features of the cleaning plate body comprise a reflective duty ratio, a reflective rectangular degree and global uniformity; extracting a reflection area in a roof photovoltaic panel image through an HSV color space threshold value, and calculating the ratio of the number of reflection pixels to the total number of pixels to serve as a reflection duty ratio; The process for obtaining the reflection rectangle comprises the steps of obtaining the minimum circumscribed rectangle for the outline of the reflection area, calculating the average value ratio of the outline area to the circumscribed rectangle area, and taking the average value ratio as the reflection rectangle; The global uniformity obtaining process comprises the step of calculating the standard deviation reciprocal of the gray level diagram after denoising the roof photovoltaic panel image, and taking the standard deviation reciprocal as the global uniformity.
3. The method for recognizing the roof photovoltaic abnormality based on the multi-dimensional exclusive feature according to claim 1, wherein in the process of extracting the bird dung shielding exclusive feature, firstly, an HSV mask and a YCbCr mask in a roof photovoltaic panel image are extracted, and the HSV mask and the YCbCr mask are subjected to AND operation to obtain a bird dung shielding candidate area mask, so that the bird dung shielding exclusive feature is extracted.
4. A method of identifying rooftop photovoltaic anomalies based on multidimensional proprietary features as recited in claim 3, wherein the bird's droppings-blocking proprietary features include Cr effective duty cycle, cr peak position, cr-Cb difference mean, edge compactness, saturation fluctuation, and area duty cycle; Counting the number of pixels with Cr values of 100-140 in a bird dung shielding candidate area, and calculating the ratio of the number of pixels to the total effective number of pixels of a mask in the bird dung shielding candidate area to obtain the effective Cr ratio; Calculating a histogram of a Cr channel region of a bird dropper shielding candidate region, and taking a Cr value corresponding to a histogram peak value as a Cr peak value position; the acquisition process of the Cr-Cb difference value average value comprises the steps of calculating the average value of a bird dung shielding candidate area as the Cr-Cb difference value average value; extracting contours of the bird dung shielding candidate areas, calculating the ratio of the area of each contour to the area of the corresponding convex hull, and taking an average value to obtain the edge compactness; The acquisition process of the saturation fluctuation comprises the steps of carrying out Gaussian blur processing on an S channel of an HSV mask, and calculating an absolute difference mean value of an original S channel and a blurred S channel in a bird droppings shielding candidate area to obtain the saturation fluctuation; The obtaining process of the area occupation ratio comprises the steps of calculating the ratio of the effective pixel number in the bird droppings shielding candidate area to the total pixel number of the image to obtain the area occupation ratio.
5. The method for recognizing the roof photovoltaic abnormality based on the multi-dimensional proprietary feature according to claim 3, wherein in the process of extracting the dust pollution proprietary feature, an initial dust pollution area is firstly extracted from the roof photovoltaic panel image through a V-channel threshold, and then non-operation and AND operation are performed on the mask of the initial dust pollution area and the mask of the bird droppings shielding candidate area, so as to obtain a purified dust pollution area, thereby extracting the dust pollution proprietary feature.
6. The method for identifying the photovoltaic anomalies on the roof based on the multi-dimensional special features according to claim 5, wherein the special features for dust pollution comprise distribution entropy, local contrast and gray standard deviation; Dividing a roof photovoltaic panel image into grids, calculating the duty ratio of dust areas in each grid, and quantifying the dust dispersion degree through an information entropy formula to obtain a distribution entropy; The local contrast obtaining process comprises the steps of calculating a contrast mean value of the dust pollution area by adopting an 8-neighborhood contrast operator to serve as local contrast; The gray standard deviation obtaining process comprises the step of calculating the standard deviation of the gray value after denoising in the dust area, and taking the standard deviation as the gray standard deviation.
7. The method for identifying the photovoltaic abnormality of the roof based on the multi-dimensional proprietary feature of claim 1, wherein the body rupture specific features include edge density and aspect ratio; The edge density obtaining process comprises the steps of extracting crack edges by Canny edge detection on a roof photovoltaic panel image, and calculating the ratio of the number of edge pixels to the total number of pixels to be used as the edge density; The obtaining process of the length-width ratio comprises the steps of solving the minimum circumscribed rectangle for the crack edge, and calculating the average value of the length-width ratio of the minimum circumscribed rectangle to obtain the length-width ratio.
8. The method of claim 1, wherein the common assist features include LBP energy, LBP entropy, GLCM contrast, GLCM correlation, GLCM energy, hsv_s mean, hsv_v mean, ycbcr_cr mean, and ycbcr_cb mean.
9. The roof photovoltaic anomaly identification method based on the multi-dimensional exclusive features of claim 1 is characterized in that the machine learning model is XGBoost model, the class weight of the XGBoost model is calculated based on a policy of balance, and the weight of a bird dung shielding class is improved on the basis of initial weight.
10. A roof photovoltaic anomaly identification system based on multi-dimensional proprietary features, comprising a memory and a processor, the memory storing a computer program, the processor invoking the computer program to perform the steps of the method of any one of claims 1-9.

Description

Roof photovoltaic anomaly identification method and system based on multidimensional proprietary features Technical Field The invention relates to the technical field of roof photovoltaic anomaly identification, in particular to a roof photovoltaic anomaly identification method and system based on multidimensional special characteristics. Background With the wide application of photovoltaic industry in urban roof scenes, roof photovoltaic operation faces two major core challenges: Firstly, due to the influence of urban complex environments (such as frequent bird activities, rapid dust accumulation and changeable illumination conditions), the photovoltaic power generation panel is easy to attach various foreign matters and has abundant characteristics, and the targeted defect eliminating means is selected to depend on the foreign matter classifying accuracy, so that the machine image processing technology is difficult to understand the events of the same foreign matters and the foreign matters, the pollution and shielding rule is difficult to perceive, and the automation degree of the foreign matter defect eliminating technology is restricted; Secondly, traditional manual inspection mode inefficiency, with high costs, and easily lead to unusual misjudgement or misjudgement because of subjective judgement, be difficult to satisfy extra large city roof photovoltaic "safe, high-efficient, unmanned" operation and maintenance demand. The existing photovoltaic anomaly identification technology mostly adopts a single characteristic (such as a gray threshold value and simple textures) or a general deep learning model, wherein the single characteristic cannot distinguish similar anomaly types (such as light birds shielding and cleaning plate reflecting areas), the identification accuracy is low, and the general deep learning model is difficult to deploy on a roof photovoltaic edge monitoring terminal because a large amount of computing resources are needed. Therefore, the invention provides a photovoltaic panel foreign matter feature extraction and classification method integrating the idea of a multi-learner decision tree so as to match the operation and maintenance requirements of a roof photovoltaic complex environment. Disclosure of Invention The invention aims to overcome the defects of the prior art and provide a roof photovoltaic anomaly intelligent recognition method based on the multidimensional special characteristics and the lightweight machine learning model, so that the high-precision recognition of four anomalies, namely cleaning a plate body, shielding by bird droppings, polluting dust, and cracking the plate body, is realized, and the low resource consumption requirement of a roof photovoltaic edge monitoring terminal is met. The aim of the invention can be achieved by the following technical scheme: a roof photovoltaic anomaly identification method based on multidimensional proprietary features comprises the following steps: acquiring a roof photovoltaic panel image, and preprocessing the roof photovoltaic panel image; Carrying out multi-dimensional feature extraction on the preprocessed roof photovoltaic panel image to obtain multi-dimensional feature vectors, wherein the multi-dimensional features comprise cleaning panel body exclusive features, bird dung shielding exclusive features, dust pollution exclusive features, panel body cracking exclusive features and general auxiliary features; and carrying out roof photovoltaic anomaly identification according to the multidimensional feature vector by adopting a machine learning model. Further, the special characteristics of the cleaning plate body comprise a light reflection ratio, a light reflection rectangular degree and global uniformity; extracting a reflection area in a roof photovoltaic panel image through an HSV color space threshold value, and calculating the ratio of the number of reflection pixels to the total number of pixels to serve as a reflection duty ratio; The process for obtaining the reflection rectangle comprises the steps of obtaining the minimum circumscribed rectangle for the outline of the reflection area, calculating the average value ratio of the outline area to the circumscribed rectangle area, and taking the average value ratio as the reflection rectangle; The global uniformity obtaining process comprises the step of calculating the standard deviation reciprocal of the gray level diagram after denoising the roof photovoltaic panel image, and taking the standard deviation reciprocal as the global uniformity. Further, in the extraction process of the special feature of the bird dung shielding, firstly, an HSV mask and a YCbCr mask in a roof photovoltaic panel image are extracted, and the HSV mask and the YCbCr mask are subjected to AND operation to obtain a candidate region mask of the bird dung shielding, so that the special feature of the bird dung shielding is extracted. Further, the special characteristics of the bird dung