Solar cell defect detection is crucial for quality inspection in photovoltaic power generation modules. In the production process, defect samples occur infrequently and exhibit random shapes and sizes, which makes it challenging to collect defective samples. Additionally, the complex surface background of polysilicon cell wafers complicates the accurate identification and localization of defective regions. This paper proposes a novel Lightweight Multi-scale Feature Fusion network (LMFF) to address these challenges. The network comprises a feature extraction network, a multi-scale feature fusion module (MFF), and a segmentation network. Specifically, a feature extraction network is proposed to obtain multi-scale feature outputs, and a multi-scale feature fusion module (MFF) is used to fuse multi-scale feature information effectively. In order to capture finer-grained multi-scale information from the fusion features, we propose a multi-scale attention module (MSA) in the segmentation network to enhance the network’s ability for small target detection. Moreover, depthwise separable convolutions are introduced to construct depthwise separable residual blocks (DSR) to reduce the model’s parameter number. Finally, to validate the proposed method’s defect segmentation and localization performance, we constructed three solar cell defect detection datasets: SolarCells, SolarCells-S, and PVEL-S. SolarCells and SolarCells-S are monocrystalline silicon datasets, and PVEL-S is a polycrystalline silicon dataset. Experimental results show that the IOU of our method on these three datasets can reach 68.5%, 51.0%, and 92.7%, respectively, and the F1-Score can reach 81.3%, 67.5%, and 96.2%, respectively, which surpasses other commonly used methods and verifies the effectiveness of our LMFF network.

To address the challenges of significant scale variations and difficulty in extracting discriminative features for small targets in airborne remote sensing object detection, while considering the constraints and efficiency requirements of onboard systems, this article proposes a lightweight detection algorithm. First, a structural reparameterization strategy is applied to optimize depthwise separable convolutions, simplifying complex structures during the inference stage, which improves inference speed and memory utilization. Second, the G-Shuffle module is designed to significantly enhance feature extraction efficiency and interchannel information interaction, balancing computational complexity and detection accuracy. Lastly, a dynamic multiscale pyramid network is introduced, employing pyramid convolution to effectively extract and fuse multiscale features. Additionally, a channel attention network is integrated to establish channel interconnections, expanding the receptive field and enhancing the extraction of finegrained features at different scales, thereby improving the detection capability for targets of various sizes, especially small targets. Experimental results demonstrate that the proposed method achieves a mAP@0.5 of 0.552 and mAP@0.5:0.95 of 0.265 on the AI-TOD dataset, with a 2.7 ms improvement in inference speed and a 14% reduction in model parameters. These results indicate superior performance across multiple metrics compared to existing state-of-the-art methods, proving the method's potential and practicality for airborne remote sensing object detection.

Unsupervised methods based on density representation have shown their abilities in anomaly detection, but detection performance still needs to be improved. Specifically, approaches using normalizing flows can accurately evaluate sample distributions, mapping normal features to the normal distribution and anomalous features outside it. Consequently, this paper proposes a Normalizing Flow-based Bidirectional Mapping Residual Network (NF-BMR). It utilizes pre-trained Convolutional Neural Networks (CNN) and normalizing flows to construct discriminative source and target domain feature spaces. Additionally, to better learn feature information in both domain spaces, we propose the Bidirectional Mapping Residual Network (BMR), which maps sample features to these two spaces for anomaly detection. The two detection spaces effectively complement each other’s deficiencies and provide a comprehensive feature evaluation from two perspectives, which leads to the improvement of detection performance. Comparative experimental results on the MVTec AD and DAGM datasets against the Bidirectional Pre-trained Feature Mapping Network (B-PFM) and other state-of-the-art methods demonstrate that the proposed approach achieves superior performance. On the MVTec AD dataset, NF-BMR achieves an average AUROC of 98.7% for all 15 categories. Especially, it achieves 100% optimal detection performance in five categories. On the DAGM dataset, the average AUROC across ten categories is 98.7%, which is very close to supervised methods.

KSII Transactions on Internet and Information Systems TIIS Paper Details

Video Anomaly Detection (VAD), aiming to identify abnormalities within a specific context and timeframe, is crucial for intelligent Video Surveillance Systems. While recent deep learning-based VAD models have shown promising results by generating high-resolution frames, they often lack competence in preserving detailed spatial and temporal coherence in video frames. To tackle this issue, we propose a self-supervised learning approach for VAD through an inter-patch relationship prediction task. Specifically, we introduce a two-branch vision transformer network designed to capture deep visual features of video frames, which can address spatial and temporal dimensions responsible for modeling appearance and motion patterns, respectively. The inter-patch relationship in each dimension is decoupled into inter-patch similarity and the order information of each patch. To mitigate memory consumption, we convert the order information prediction task into a multi-label learning problem, and the inter-patch similarity prediction task into a inter-patch distance matrix regression problem. Comprehensive experiments demonstrate the effectiveness of our method, surpassing pixel-generation-based methods by a significant margin across three public benchmarks. Additionally, our approach outperforms other self-supervised learning-based methods.