Small Object Recognition Algorithm of Grain Pests Based on SSD Feature Fusion

The detection of grain pests is great significance to storage. However, in practice, because the size insects too small identify. In this paper, feature fusion SSD(single shot multi-box detector) algorithm based on Top-Down strategy was proposed. Firstly, module used fuse output characteristics conv4 and conv5, block 11 which not conducive object deleted. Secondly, K-means clustering cluster prior bounding boxes made them more suitable for pests, improves performance pests. Five methods were enhance self-made dataset enhanced reached 9990 images. Experiments show that optimized model achieves a mAP (mean Average Precision) 96.89% with speed 0.040s per image. Compared 95.45% achieved by original SSD algorithm, proposed has improvement performance. two-stages Faster R-CNN (mAP 90.53% 0.115s image), YOLOv3, TDFSSD EfficentDet (D2, D1), accuracy have obvious advantages. experimental results good certain guiding subsequent pest image detection.