Research On Improved Algorithm Of Object Detection Based On SSD

SSD(Single Shot MultiBox Detector)is a popular target detection method.At present,there are many target detection methods.Target detection using convolutional neural networks occupies a dominant position.However,convolutional neural networks have inherent problems in structure:high-level networks have large receptive fields and semantic information has strong ability to represent,but the resolution is low,geometric detail information is weak.The low-level network has relatively small receptive fields,and it has strong geometric detail information representation capability.Although the resolution is high,the semantic information representation ability is weak.The SSD uses multi-scale feature mAPs to predict objects,uses high-level feature information with large receptive fields to predict large objects,and has low receptive fields for low-level feature information to predict small objects.This brings a problem:When using low-level network feature information to predict small objects,due to the lack of high-level semantic features,SSD have a poor detection effect on small objects.Based on the analysis and introduction of classic SSD algorithms,this paper proposes two improved algorithms for the existing problems of the newer SSD algorithm.1.MSSD(Modified Single Shot MultiBox Detector)is proposed.This article uses the FPN-based network architecture to integrate high and low layers and improves the traditionally sampled structure.The high-level semantic information is integrated into the low-level network feature information,and the multi-scale feature mAPs for predicting the regression location box and the classification task input are enriched to improve the detection accuracy.The VGG16 network used by the SSD training is replaced with a deep residual network to optimize candidate box regression and classification task input feature mAPs to improve detection accuracy and detection speed.Experiments show that the MSSD model is superior to the traditional SSD model both in detection accuracy and detection speed.2.A TMSSD(Top-Down Single Shot MultiBox Detector)model is proposed.We improve the network speed by modifying the number of channels in the prediction layer.This paper improves the network speed by optimizing the channel of the prediction layer and we improved ResNet.It adopts the upsampling method used in the TDM(Top-Down Modulation)structure.The feature pyramid network structure in MSSD is improved,high-level semantic information is integrated with low-level semantic information,and detection accuracy is improved.Experiments show that the TMSSD model is superior to the MSSD model both in detection accuracy and detection speed.