A mobilenet-SSD target detection algorithm based on attention mechanism

Abstract

Object detection based on convolutional neural network has important applications in mobile devices, intelligent robots and other fields. However, due to the diversity of traditional models, large number of parameters and slow computing speed, it is difficult to meet the real-time detection requirements of embedded systems. To solve this problem, a mobilenet-SSD model integrating attention mechanism was proposed. Based on the lightweight network model Mobilenet-SSD, the convolution block attention module was integrated into the model, which greatly improved the accuracy of the detection model at the cost of a small increase in computation. Meanwhile, CIOU loss function is used as position loss function to optimize learning strategy and improve training effect. In addition, the processing mode of pre-selection box is improved to Soft-NMS to optimize the detection performance of multi-target overlapping. Experimental results on Pascal VOC2007 and VOC2012 data sets show that the proposed model achieves good performance in recognition accuracy and detection speed, with detection speed of 60.9 FPS and mAP accuracy of 73.6% on GTX1080.The model in this paper has the advantages of small number of parameters, fast computing speed and high identification accuracy, which can better meet the real-time detection requirements of mobile terminals and embedded systems, and provide a good target detection scheme for industrial systems and portable systems.