The performance of the activation function in convolutional neural networks is directly related to the model’s image classification accuracy. The rectified linear unit (ReLU) activation function has been extensively used in image classification models but has significant shortcomings, including low classification accuracy. The performance of a series of parametric activation functions has made parameter addition a popular research avenue for improving the performance of activation functions in recent years, and excellent progress has been achieved. Existing parametric activation functions often focus on assigning a different slope to the negative part of the activation function and still involve the negative value alone in the activation function calculation, without considering the impact of linking the negative value to the positive value on the performance of the activation function. As a result, this work proposes a novel parametric right-shift activation function, the adaptive offset activation function (AOAF). By inserting an adaptive parameter (the mean value of the input feature tensor) and two custom ReLU parameters, the negative parameters previously driven to zero by ReLU can be turned into positive parameters with lower weight and participate in CNN feature extraction. We compared the performance of the suggested activation function to the performance of a selection of typical activation functions using four distinct public datasets. Compared with ReLU, the average classification accuracy of our proposed activation function improved by 3.82%, 0.6%, 1.02%, and 4.8% for the four datasets, respectively.

Multi-level thresholding image segmentation divides an image into multiple regions of interest and is a key step in image processing and image analysis. Aiming toward the problems of the low segmentation accuracy and slow convergence speed of traditional multi-level threshold image segmentation methods, in this paper, we present multi-level thresholding image segmentation based on an improved slime mould algorithm (ISMA) and symmetric cross-entropy for global optimization and image segmentation tasks. First, elite opposition-based learning (EOBL) was used to improve the quality and diversity of the initial population and accelerate the convergence speed. The adaptive probability threshold was used to adjust the selection probability of the slime mould to enhance the ability of the algorithm to jump out of the local optimum. The historical leader strategy, which selects the optimal historical information as the leader for the position update, was found to improve the convergence accuracy. Subsequently, 14 benchmark functions were used to evaluate the performance of ISMA, comparing it with other well-known algorithms in terms of the optimization accuracy, convergence speed, and significant differences. Subsequently, we tested the segmentation quality of the method proposed in this paper on eight grayscale images and compared it with other image segmentation criteria and well-known algorithms. The experimental metrics include the average fitness (mean), standard deviation (std), peak signal to noise ratio (PSNR), structure similarity index (SSIM), and feature similarity index (FSIM), which we utilized to evaluate the quality of the segmentation. The experimental results demonstrated that the improved slime mould algorithm is superior to the other compared algorithms, and multi-level thresholding image segmentation based on the improved slime mould algorithm and symmetric cross-entropy can be effectively applied to the task of multi-level threshold image segmentation.

To resolve the problems associated with the small target presented by printed circuit board surface defects and the low detection accuracy of these defects, the printed circuit board surface-defect detection network DCR-YOLO is designed to meet the premise of real-time detection speed and effectively improve the detection accuracy. Firstly, the backbone feature extraction network DCR-backbone, which consists of two CR residual blocks and one common residual block, is used for small-target defect extraction on printed circuit boards. Secondly, the SDDT-FPN feature fusion module is responsible for the fusion of high-level features to low-level features while enhancing feature fusion for the feature fusion layer, where the small-target prediction head YOLO Head-P3 is located, to further enhance the low-level feature representation. The PCR module enhances the feature fusion mechanism between the backbone feature extraction network and the SDDT-FPN feature fusion module at different scales of feature layers. The C₅ECA module is responsible for adaptive adjustment of feature weights and adaptive attention to the requirements of small-target defect information, further enhancing the adaptive feature extraction capability of the feature fusion module. Finally, three YOLO-Heads are responsible for predicting small-target defects for different scales. Experiments show that the DCR-YOLO network model detection map reaches 98.58%; the model size is 7.73 MB, which meets the lightweight requirement; and the detection speed reaches 103.15 fps, which meets the application requirements for real-time detection of small-target defects.

In response to the problems of low accuracy and poor noise immunity of the traditional fault diagnosis method for rolling bearing fault diagnosis due to the complex and variable operating conditions of rolling bearings and the large noise interference during bearing signal acquisition, a rolling bearing fault diagnosis model based on VMD–RP–CSRN is proposed. Firstly, the initial feature extraction of the bearing signal is carried out by variational modal decomposition (VMD), which is then converted into a two-dimensional image with fault features by recurrent plot (RP) coding, and then the feature images are input to a channel split residual network (CSRN) for feature extraction and fault classification. In order to verify the accuracy and noise immunity of the proposed method for the diagnosis of bearing faults under complex working conditions, experiments on the selection of parameters in the CSRN model were conducted on the bearing dataset of Jiangnan University, and experiments on the diagnosis of bearing faults under complex working conditions and noise immunity of CSRN were carried out and compared with other commonly used methods. The proposed bearing fault diagnosis method based on VMD–RP–CSRN combines VMD and RP to retain the fault features in the original signal to the maximum extent and stress the hidden features in the signal. The proposed channel split operation realizes the extraction of hidden features by selecting the main operating channel of the three-channel feature image, and makes more fault features participate in the feature extraction of the diagnosis model. The experimental results demonstrate that the proposed method is at least 1.2% better than the comparison method, and has better noise immunity. In addition, experiments on the fault diagnosis capability of the model with different data set sizes and the diagnosis of variable speed bearing data by the model show that the proposed method has better generalization performance and diagnosis capability.