No abstract

The beetle antenna search algorithm (BAS) converges rapidly and runs in a short time, but it is prone to yielding values corresponding to local extrema when dealing with high-dimensional problems, and its optimization result is unstable. The artificial fish swarm algorithm (AFS) can achieve good convergence in the early stage, but it suffers from slow convergence speed and low optimization accuracy in the later stage. Therefore, this paper combines the two algorithms according to their respective characteristics and proposes a mutation and a multi-step detection strategy to improve the BAS algorithm and raise its optimization accuracy. To verify the performance of the hybrid composed of the AFS and BAS algorithms based on the Mutation and Multi-step detection Strategy (MMSBAS), AFS-MMSBAS is compared with AFS, the Multi-direction Detection Beetle Antenna Search (MDBAS) Algorithm, and the hybrid algorithm composed of the two (AFS-MDBAS). The experimental results show that, with respect to high-dimensional problems: (1) the AFS-MMSBAS algorithm is not only more stable than the MDBAS algorithm, but it is also faster in terms of convergence and operation than the AFS algorithm, and (2) it has a higher optimization capacity than the two algorithms and their hybrid algorithm.

The detection of small objects is easily affected by background information, and a lack of context information makes detection difficult. Therefore, small object detection has become an extremely challenging task. Based on the above problems, we proposed a Single-Shot MultiBox Detector with an attention mechanism and dilated convolution (ADSSD). In the attention module, we strengthened the connection between information in space and channels while using cross-layer connections to accelerate training. In the multi-branch dilated convolution module, we combined three expansion convolutions with different dilated ratios to obtain multi-scale context information and used hierarchical feature fusion to reduce the gridding effect. The results show that on PASCAL VOC2007 and VOC2012 datasets, our 300 × 300 input ADSSD model reaches 78.4% mAP and 76.1% mAP. The results outperform those of SSD and other advanced detectors; the effect of some small object detection is significantly improved. Moreover, the performance of the ADSSD in object detection affected by factors such as dense occlusion is better than that of the traditional SSD.