The variousness as well as the inaccessibility of defect characteristics make the defect detection of monocrystalline silicon solar cells more challenging. To address these problems, a novel domain adaptive target detection algorithm based on pseudo-label learning, which is an efficient and feasible weakly supervised learning method, is proposed in this paper. Firstly, in the early stage of the model, the loss function is improved to solve the problem that the model is not easy to converge due to a large number of false labels in the pseudo labels. Then, to classify the sample space better, entropy regularization is applied to the sample boundary data, and the unlabels of the target domain images are labeled with pseudo labels. Finally, the source domain data and the target domain data are used for training together, and the generalization performance of the model obtained is greatly improved. The results show that in solar cell image detection, the accuracy of the domain adaptive method based on pseudo-label learning can reach over 90%, which is better than the target detection accuracy of using only the source domain dataset.
Stability detection of melt pool has become a challenging task in laser cladding, due to the high temperature and brightness during laser cladding. In this paper, an enhanced mask R-CNN for object instance segmentation of melt pool is proposed to boost the performance of detection. In order to enrich the dataset and improve the generalization of the neural network, the data enhancement method of elastic deformations is used to simulate the irregular deformation of melt pool topography caused by the interference of the external environment. Meanwhile, the MobilenetV2 structure is introduced into mask R-CNN to solve the problem of a large number of parameters and slow running speed of network model, and transformer model was used to replace the classifier of the original network. Experimental results show that the proposed method can improve testing speed by 14.7% without decreasing the segmentation accuracy. Finally, a dynamic stability detection method of melt pool is proposed in this paper.
Defect detection methods for photovoltaic (PV) devices based on electroluminescence (EL) imaging technology are crucial to maintain productivity and prolong components’ life. However, due to the lack of feature extraction capability for morphologically complex defects and some small defects, the traditional object detection algorithms perform not well in EL defect detection. Therefore, an improved algorithm based on the Faster Region-based Convolutional Network algorithm (Faster R-CNN) is proposed to improve the detection performance of multi-scale defects. Specifically, an improved residual module based on deformable convolution and attention module is proposed to improve the detection rate of morphologically complex defects. And a Feature Pyramid Network (FPN) is utilized in the proposed algorithm to improve the detection performance of small defects. In addition, the GIOU loss, instead of the original smooth L1 loss, is utilized to improve the boundary box regression accuracy. Experimental results on the detection of the EL defects show the high efficiency of the proposed algorithm. The method is expected to provide more guiding feedback in both practical design and reliable diagnosis of the PV industry.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.