Deep learning has been an important topic in fault diagnosis of motor bearings, which can avoid the need for extensive domain expertise and cumbersome artificial feature extraction. However, existing neural networks have low fault recognition rates and low adaptability under variable load conditions. In order to solve these problems, we propose a one-dimensional fusion neural network (OFNN), which combines Adaptive one-dimensional Convolution Neural Networks with Wide Kernel (ACNN-W) and Dempster-Shafer (D-S) evidence theory. Firstly, the original vibration time-domain signals of a motor bearing acquired by two sensors are resampled. Then, four frameworks of ACNN-W optimized by RMSprop are utilized to learn features adaptively and pre-classify them with Softmax classifiers. Finally, the D-S evidence theory is used to comprehensively determine the class vector output by the Softmax classifiers to achieve fault detection of the bearing. The proposed method adapts to different load conditions by incorporating complementary or conflicting evidences from different sensors through experiments on the Case Western Reserve University (CWRU) motor bearing database. Experimental results show that the proposed method can effectively enhance the cross-domain adaptive ability of the model and has a better diagnostic accuracy than other existing experimental methods.

In response to the challenge of easily falling into local optima and slow identification speed in the parameter identification of permanent magnet synchronous motors (PMSMs), this paper proposes a CGCRAO algorithm based on chaotic initialization and a hybrid variation strategy. The algorithm uses Tent chaotic mapping for population initialization to improve population diversity. At the same time, by combining the Gaussian and Cauchy distribution characteristics and the three-stage operation idea, the optimal individual variation strategy is autonomously selected in real time to improve the RAO-1 algorithm. This paper validates the effectiveness of the algorithm improvement and the correctness of the three-stage operation idea using eight benchmark test functions. Furthermore, This paper conducted comparative experiments on parameter identification of five algorithms under different operating conditions through simulation and experiments. The results indicate that the proposed CGCRAO algorithm enables fast and accurate identification of PMSM parameters.

The fixed-pattern noise (FPN) caused by nonuniform optoelectronic response limits the sensitivity of an infrared imaging system and severely reduces the image quality. Therefore, nonuniform correction of infrared images is very important. In this paper, we propose a deep filter neural network to solve the problems of network underfitting and complex training with convolutional neural network (CNN) applications in nonuniform correction. Our work is mainly based on the idea of deep learning, where the nonuniform image noise features are fully learned from a large number of simulated training images. The network is designed by introducing the filter and the subtraction structure. The background interference of the image is removed by the filter, so the learning model is gathered in the nonuniform noise. The subtraction structure is used to further reduce the input-to-output mapping range, which effectively simplifies the training process. The results from the test on infrared images shows that our algorithm is superior to the state-of-the-art algorithm in visual effects and quantitative measurements, providing a new method for deep learning in nonuniformity correction of single images.

In order to identify the parameters of photovoltaic (PV) cells and modules more accurately, reliably and efficiently, a chaotic second order oscillation JAYA algorithm (CSOOJAYA) is proposed. Firstly, both logical chaotic map and the mutation strategy are brought in enhancing the population diversity and improving exploitation. Secondly, in order to better balance exploration and exploitation, the second order oscillation factor is added, which not only improves the diversity of the population, but also has strong exploration at the beginning of the iteration and strong exploitation at the end of the iteration. In order to balance the two abilities, the self-adaptive weight is introduced into the CSOOJAYA algorithm to regulate individuals the tendency of moving toward the optimal solution and escaping from the worst solution, so as to enhance the search efficiency and exploitation. In order to validate the behavior of CSOOJAYA, it is employed to the parameter identification problem of PV models. Finally, the experimental results show that CSOOJAYA delivers excellent behavior in the aspects of convergence, reliability and accuracy.

Photovoltaic (PV) technology plays a pivotal role in the conversion of solar energy into electricity. However, subtle defects in solar cells within PV modules can significantly impact both the energy efficiency and the lifespan of these modules. Accurate identification of such defects is essential to ensure measurement reliability and maintain production quality. Traditional electroluminescence (EL) image-based detection methods often encounter limitations, including inadequate multiscale feature extraction, low sensitivity to minor defects, and excessive model complexity. To address these challenges, we propose the Scale Adaptive Feature Integration Network (SAFINet), an enhanced detection model. SAFINet integrates a complex intercropping feature pyramid network (CIFPN) to strengthen inter-layer feature interactions and employs a Multi-scale Feature Aggregation Attention (MSFAA) mechanism to enhance feature representation across various scales. Furthermore, DualConv modules replace standard convolutions to optimize efficiency, while a Scale-adaptive Weighted Intersection over Union (SA-WIoU) is introduced as a regression loss function aimed at accelerating convergence and improving localization accuracy. Evaluated on the PVEL-AD and ELPV datasets, SAFINet demonstrates significant performance improvements. On the PVEL-AD dataset, the model achieves mean Average Precision scores of mAP_0.5 at 98.8% and mAP_0.5-0.95 at 81.7%, reflecting enhancements of 5.1% and 7.5% over the baseline, respectively. In addition, the inference speed increases from 118 FPS to 178 FPS. In testing on ELPV data, mAP_0.5 improves by 7.2%, while mAP_0.5-0.95 sees an increase of 11.1%, with FPS increasing from 84 to 119. These results confirm the effectiveness of SAFINet in achieving an optimal balance between defect detection accuracy and processing speed, providing high practical value for the diagnosis of PV defects and quality assurance.

Accurate object detection plays a critical role in image processing, especially for identifying small targets in complex backgrounds. However, factors such as degraded image quality and extended viewing distances impede detection performance, compromising both accuracy and multi-target recognition. To address these challenges, we propose the Multi-Attention Spatial Network (MASNet), a streamlined object detection model that leverages attention mechanisms to enhance target features while suppressing noise interference in cluttered environments. MASNet introduces Space-to-Depth Convolution (SPDConv) to enrich the feature map depth while preserving spatial resolution, thereby improving local detail extraction and reducing computational overhead. Additionally, Global-Local Spatial Attention (GLSA) is integrated into the backbone to emphasize critical features across both global and local scales, enhancing fine-grained feature representation. Furthermore, Sparse Multi-Scale Attention (SMSA) is incorporated into the detection head to refine multi-scale feature learning, while the Efficient IoU (EIoU) loss function is employed to accelerate convergence, improve regression accuracy, and boost overall detection performance. Experimental results demonstrate that MASNet achieves a mAP of 52.2 % on the VisDrone2019 test set and 62.8 % on the validation set, significantly enhancing detection accuracy for aerial images. Moreover, MASNet attains a mAP of 89.7 % on the AFO dataset, 42.8 % on the UAVDT dataset, and 85.1 % on the UAVOD-10 dataset. We confirmed MASNet’s superior performance through comprehensive comparative evaluations across multiple datasets, demonstrating its robustness and efficiency in small target detection under diverse and challenging imaging conditions.