A Mini-UAV Lightweight Target Detection Model Based on SSD

Mini-UAV can not carry high-performance computing equipment, and the conventional neural network model is difficult to deploy to Mini-UAV because of its large scale and complex calculation. To solve the problem of the huge amount of computation of the deep learning model, we introduce a lightweight object detection network model for Mini-UAV that greatly reduces the amount of model computation and parameters on the premise of ensuring the detection accuracy. In this paper, SSD is used as the benchmark object detection model, depthwise separable convolution and grouped convolution are used as the basic lightweight means. A simplified grouped heterogeneous convolution structure is introduced and a spatial/channel hybrid attention mechanism is also introduced to achieve high-lowlayer feature fusion. Pascal VOC 2012 dataset is used for training and testing. We compared our algorithm with various lightweight target detection models from the perspective of model accuracy and model size. The experimental comparison results show that our model can improve detection accuracy with a lower computational cost.