Neural network pruning, an important method to reduce the computational complexity of deep models, can be well applied to devices with limited resources. However, most current methods focus on some kind of information about the filter itself to prune the network, rarely exploring the relationship between the feature maps and the filters. In this paper, two novel pruning methods are proposed. First, a new pruning method is proposed, which reflects the importance of filters by exploring the information in the feature maps. Based on the premise that the more information there is, more important the feature map is, the information entropy of feature maps is used to measure information, which is used to evaluate the importance of each filter in the current layer. Further, normalization is used to realize cross layer comparison. As a result, based on the method mentioned above, the network structure is efficiently pruned while its performance is well reserved. Second, we proposed a parallel pruning method using the combination of our pruning method above and slimming pruning method which has better results in terms of computational cost. Our methods perform better in terms of accuracy, parameters, and FLOPs compared to most advanced methods. On ImageNet, it is achieved 72.02% top1 accuracy for ResNet50 with merely 11.41M parameters and 1.12B FLOPs.For DenseNet40, it is obtained 94.04% accuracy with only 0.38M parameters and 110.72M FLOPs on CIFAR10, and our parallel pruning method makes the parameters and FLOPs are just 0.37M and 100.12M, respectively, with little loss of accuracy.

In this paper, we propose a new visual tracking method in light of salience information and deep learning. Salience detection is used to exploit features with salient information of the image. Complicated representations of image features can be gained by the function of every layer in convolution neural network (CNN). The characteristic of biology vision in attention-based salience is similar to the neuroscience features of convolution neural network. This motivates us to improve the representation ability of CNN with functions of salience detection. We adopt the fully-convolution networks (FCNs) to perform salience detection. We take parts of the network structure to perform salience extraction, which promotes the classification ability of the model. The network we propose shows great performance in tracking with the salient information. Compared with other excellent algorithms, our algorithm can track the target better in the open tracking datasets. We realize the 0.5592 accuracy on visual object tracking 2015 (VOT15） dataset. For unmanned aerial vehicle 123 (UAV123) dataset, the precision and success rate of our tracker is 0.710 and 0.429.

Single-photon LiDAR possesses weak echo sensing capability, but random backgrounds can easily cause target response loss and fragmented false alarms. In particular, sparse return photons, background photons, and dark counts may generate stochastic responses, making weak target preservation and false-alarm suppression difficult under single-photon observations. This paper proposes STMP-UNet, which uses a structure-aware target alignment module to suppress ineffective shallow response propagation through decoder semantic constraints, and combines a morphology prior module to support the spatial continuity of coarse target-region responses. The former selectively transfers target-related shallow information under high-level semantic guidance, while the latter introduces multi-scale structural constraints to alleviate spatial fragmentation of sparse target responses. Experiments are conducted on simulated single-photon data constructed from Anti-UAV infrared sequences. A photon-counting statistical simulation is used to reproduce sparse photon responses, and an infrared-derived proximity map provides auxiliary spatial structural information. Results show that compared to Pure U-Net, STMP-UNet improves Object F1 from 0.7115 to 0.7531 and Pd from 0.6635 to 0.7074, and reduces FPPI from 0.1907 to 0.1619. Ablation and candidate-level FROC analyses further demonstrate the effectiveness of the proposed structure modeling for preserving weak target responses and improving the detection–false-alarm trade-off.