Small object detection represents a pivotal sub-domain within the field of computer vision. Previous research aimed at enhancing detection accuracy has included augmenting the head layer, refining multi-layer feature pooling techniques, incorporating attention mechanisms, and optimizing loss functions. Despite these efforts, issues such as false negatives and classification ambiguities persist, leading to suboptimal outcomes. To solve these issues, DKA-YOLO is proposed as a new model focusing on improving convolution kernel structures. We develop novel modules based on the concept of dilation kernels aggregation convolution, integrate them into the robust and advanced YOLOv8 framework, and apply the enhanced model to small object detection tasks. The proposed modules include the large size dilation kernels aggregation convolution for the backbone layer, which combines large kernel sizes with dilation convolution structure, and utilizes extensive receptive fields to improve detailed feature extraction. Additionally, the multi-scale dilation kernels aggregation convolution is introduced in the neck layers to enhance performance and efficiency with a diverse set of kernels. Finally, the model’s head layer employs multi-scale convolution kernels detect to enhance feature expression diversity, generalization ability, and computational efficiency of detection. Experimental validation on public datasets demonstrates a significant improvement in detection accuracy by our method, with an increase in mean average precision by 1.5% on the VisDrone and 1.15% on the UAVDT compared to advanced previous methods. Our method also surpasses other previous models in comparative experiments, highlighting its superiority and robustness.

Neural network pruning is essential for deploying deep learning models on resource-constrained devices by reducing computational and memory demands. In this paper, we propose a novel pruning framework, Entropy-Guided Search Space Optimization for Efficient Neural Network Pruning, which uses information entropy to assess the importance of convolutional layers. Specifically, we calculate the layer-wise entropy of pretrained weights, apply outlier detection to remove extreme values, and normalize the entropy values. These normalized values guide the selection of retention ratios, ensuring that layers with higher entropy retain more filters. By refining the subnetwork search space, our approach enhances the efficiency of the search process and improves overall subnetwork performance. The refined search space targets more promising regions, reducing computational overhead and resulting in higher-quality pruned networks. Through iterative optimization, the optimal subnetwork is identified and fine-tuned to produce the final pruned model. Experimental results on benchmark datasets show that our method significantly outperforms existing pruning methods, achieving substantial improvements in both accuracy and computational efficiency. This entropy-guided pruning strategy provides a robust and effective solution for neural network compression, suitable for a wide range of deep learning applications.