In the single pruning algorithm, channel pruning or filter pruning is used to compress the deep convolution neural network, and there are still many redundant parameters in the compressed model. Directly pruning the filter will largely cause the loss of key information and affect the accuracy of model classification. To solve these problems, a parallel pruning algorithm combined with image enhancement is proposed. Firstly, in order to improve the generalization ability of the model, a data enhancement method of random erasure is introduced. Secondly, according to the trained batch normalization layer scaling factor, the channels with small contribution are cut off, the model is initially thinned, and then the filters are pruned. By calculating the geometric median of the filters, redundant filters similar to them are found and pruned, and their similarity is measured by calculating the distance between filters. Pruning was done using VGG19 and DenseNet40 on cifar10 and cifar100 data sets. The experimental results show that this algorithm can improve the accuracy of the model, and at the same time, it can compress the calculation and parameters of the model to a certain extent. Finally, this method is applied in practice, and combined with transfer learning, traffic objects are classified and detected on the mobile phone.

To solve the problem of complex network models with a large number of redundant parameters, a pruning algorithm combined with an attention mechanism is proposed. Firstly, the basic training is performed once, and the network model is then re-trained with the attention mechanism for the baseline. The obtained model is pruned based on channel correlation, and finally a simplified model is obtained via continuous cyclic iteration while the accuracy rate is kept as close as possible to that of the baseline model. The algorithm was experimentally validated on ResNet based on different datasets, and the results showed that the algorithm provided strong adaptability to different datasets and different network structures. For the CIFAR-100 dataset, ResNet50 was pruned to reduce the amount of model parameters by 80.3% and the amount of computation by 69.4%, while maintaining accuracy. For the ImageNet dataset, the ResNet50 parameter volume was compressed by 2.49 times and the computational volume was compressed by 3.01 times. The ResNet101 parameter volume was reduced by 61.2%, and the computational volume was reduced by 68.5%. Compared with the traditional fixed threshold, the model achieves better results in terms of detection accuracy, compression effect, and inference speed.

In this paper, a compact cat swarm optimization algorithm based on a Small Sample Probability Model (SSPCCSO) is proposed. In the same way as with previous algorithms, there is a tracking mode and a searching mode in the processing of searching for optimal solutions, but besides these, a novel differential operator is introduced in the searching mode, and it is proved that this could greatly enhance the search ability for the potential global best solution. Another highlight of this algorithm is that the gradient descent method is adopted to increase the convergence velocity and reduce the computation cost. More importantly, a small sample probability model is designed to represent the population of samples instead of the normal probability distribution. This representation method could run with low computing power of the equipment, and the whole algorithm only uses a cat with no historical position and velocity. Therefore, it is suitable for solving optimization problems with limited hardware. In the experiment, SSPCCSO is superior to other compact evolutionary algorithms in most benchmark functions and can also perform well compared to some population-based evolutionary algorithms. It provides a new means of solving small sample optimization problems.

Retinal artery/vein segmentation is a prerequisite for many ophthalmic diagnostic tools. Yet, the task remains difficult: vessels form complex trees, vary widely in caliber, and often appear low-contrast at terminal branches. We propose SDA-SwinNet to handle these challenges. The network adopts Swin-UNet as its backbone and adds three modifications: a Shift-ASPP module for multi-scale context, an HF-Bridge for cross-level feature fusion, and a fractal-constrained loss with a differentiable topology surrogate. Experimental results on the DRIVE-AV and LES-AV datasets show that the proposed model achieves an overall F1-score of 73.13% on DRIVE-AV and 67.85% on LES-AV, with additional class-wise evaluations for arteries and veins. The results demonstrate that SDA-SwinNet achieves a competitive trade-off between segmentation accuracy and computational efficiency.