Research On Target Detection And Recognition Algorithm Based On SSD And Inception?resnet?v2 Network

Target detection and recognition is a technology for classifying and locating target objects in images.It is an important research direction in the field of computer vision.It is also the basis of many visual tasks,such as target tracking,face recognition,and autonomous driving.However,in practical applications,the scenes are often more complex.There are many problems,such as occlusion between objects,object color change,object size change and so on,which result in unsatisfactory detection effect of the algorithm.Among the current mainstream algorithms,the SSD(Single Shot multibox Detector)algorithm has better overall performance.It can take into account detection accuracy and speed at the same time,however there is still room for improvement.In order to improve the detection accuracy of the SSD algorithm,this paper proposes two improvements.(1)A network model IRSSD based on the Inception?resnet structure is proposed replace part of the convolutional layer in the VGG16 network.The sparse connection method is utilized to optimize the network structure which increases the depth and width of the network.So the model can extract more feature information and enhance the feature extraction ability of the network.(2)There are fewer low-level feature convolutional layers in the model.Feature extraction is insufficient and the detection effect is poor.Aiming at this problem,a network structure FIRSSD(Feature Fusion Inception?resnet Single Shot multibox Detector)with multi-layer feature fusion is proposed.The features in Conv9?2 are upsampled by transposed convolution and merged into Conv4?3 layer by channel splicing.And through the maximum pooling of the feature information in Conv3?3,it is integrated into Conv7 in the way of "element-sum" to improve the feature expression ability of the feature layer.Through the comparison of experimental results,it is proved that the improvement of the SSD algorithm in this paper effectively improves the feature extraction ability of the network and enhances the target detection accuracy of the algorithm.