To realize the defect detection of Photovoltaic (PV) modules based on infrared images, a one-stage detector based on FPT and loss function optimization is proposed. Firstly, ResNet50 is selected as the backbone network for feature extraction, combining with the idea of migration learning to strengthen its feature extraction ability. Besides, replacing the traditional feature pyramid with FPT, using Transformer structure help realize the interaction of the same position between different pyramid layers as well as different positions of the same layer, improving the feature fusion capability of the model. Finally, to alleviate the problem of suppression of non-optimal predictions by NMS strategy, a bipartite graph matching algorithm based on loss function minimization is designed to enhance the learning of the distribution of ground truth targets by the network. The dataset is expanded by Mosaic data enhancement method for model training, so as to achieve infrared image-based PV defect detection. Results show that the improved one-stage target detection network has a promising detection effect on PV module defects under infrared images, with the average accuracy improving by 2.20% compared with the original algorithm.
An Optimized One-Stage Detector for Autonomous PV Defect Detection
Lect. Notes Electrical Eng.
International Conference on Autonomous Unmanned Systems ; 2021 ; Changsha, China September 24, 2021 - September 26, 2021
Proceedings of 2021 International Conference on Autonomous Unmanned Systems (ICAUS 2021) ; Chapter : 168 ; 1707-1716
2022-03-18
10 pages
Article/Chapter (Book)
Electronic Resource
English
Optimized Filters for Texture Defect Detection
British Library Conference Proceedings | 2005
|OPTIMIZED MULTI-STAGE INTERMITTENT FUGITIVE EMISSION DETECTION
European Patent Office | 2023
|OPTIMIZED MULTI-STAGE INTERMITTENT FUGITIVE EMISSION DETECTION
European Patent Office | 2020
|Nondestructive detector and track defect detection device
European Patent Office | 2024
|OPTIMIZED MULTI-STAGE INTERMITTENT FUGITIVE EMISSION DETECTION
European Patent Office | 2021
|