Aircraft target detection method based on improved SSD

For the problem of omission and misdetection of small-scale aircraft targets in aerial images, an improving SSD object detection model is proposed based on SSD (Single Shot MultiBox Detector) model. Firstly, in view of the lack of semantic and detailed information in the shallow feature map in the SSD model, a feature fusion mechanism is designed to enrich the semantic and detailed information of the shallow feature layer by adding the supplementary feature layer obtained from the recursive reverse path. Then, to address the problem of the SSD model to focus on the channels and spatial information, a hybrid attention module combining channels and space is introduced to improve the overall attention ability of the model. Finally, the proportion of prior boxes is adjusted for the problem of mismatch to small-scale targets in the SSD model. The self-made aerial images data set is used for verification. The results show that the improved algorithm accuracy is 95. 7%, which is 7. 5% higher than original SSD algorithm, and the detection speed is 30. 8 FPS.