Aiming at the problem that the current target detection algorithm is difficult to run on the hardware platform with limited resources,based on the Single Shot MultiBox Detector(SSD),The GhostNet network is introduced into SSD to replace the original VGG16 backbone network,and the lightweight attention mechanism is introduced to compensate for the loss of detection accuracy brought by the lightweight model and improve the network detection capability.When tested on the same dataset,the size of the improved GC-SSD model is reduced from 92.11 M of the original SSD model to 10.90 M,and the mAP of the GC-SSD model is increased by 15.43%compared with the Ghost-SSD model.The test results show that the mAP of the improved GC-SSD model reached 92.06%,the model size is greatly reduced,and it is more suitable for mobile devices.
维普
万方数据