Photovoltaic Module Image Recognition Based On Improved Clustering Analysis

Exploitation and using of clean and renewable energy to improve the domestic energy structure is important for building smart city.Photovoltaic(PV)power generation is a good choose,and modules inspection is an important direction of PV power generation.The manual inspection is a common way but is not feasible for large-scale PV systems in practice due to low efficiency,high error rate and long assessment time.Currently inspection system based on Unmanned Aerial Vehicles(UAVs)has been proved to be a better solution in PV inspection.How to locate PV arrays in aerial image captured by UAV is one of the technical changes exist in UAV-based PV inspection.This paper proposed an improved adaptive clustering method based on k-means algorithm to orientate and acquire single PV array's edge,lots of monitoring data indicate this algorithm has a high accuracy and make the loss rate lower than 5%.The main work is as follows:This paper studies the particularity of uav aerial image recognition,and designs relevant algorithms to solve the problem.There are some differences between the processing scenes of uav aerial photography and ordinary image recognition.On the one hand,the location of photovoltaic power station is complex,and the aerial photography environment is poor,which causes more interference items of uav aerial photography.On the other hand,the uav platform has limited processing capability and needs to be closely integrated with the navigation system and ground station processing system,so there are relevant requirements for the adopted algorithm.According to these particularities,this paper presents a texture feature with color features constitutes a hybrid image characteristics,the clustering algorithm was improved to make it have better adaptability and other innovative ideas.Steps and the main methods for image processing was studied,and compared the advantages and limitations of various methods,innovative ways to incorporate some methods,complement each other,has obtained the good effect.In the image preprocessing stage,the image was distorted and the haze was removed.In the feature extraction,the color and texture features are fused.In the phase of image classification,we choose cluster analysis as the classifier,this paper proposes an improved cluster analysis method with self-tuning parameters.Clustering initial point selection and determination of the clustering number is the two major difficulties of the traditional clustering analysis method.In this paper,uses the color feature of the image to determine the distribution characteristics of the two initial parameters.A set of image recognition system for PV inspection was completed,and it was applied in the actual project,with a leakage rate of less than 5%,which improved the inspection efficiency and had certain economic benefits.