Object Detection And Recognition Based On Convolutional Neural Networks

With the improvement of computer storage computing power and the generation of massive image data,artificial intelligence has once again received wide attention from all walks of life and has become a new focus of scientific and technological competition among countries.The target detection and recognition of images are the basis and key of many applications of artificial intelligence.Target detection is one of the core issues in the field of machine vision.it is designed to find all the object hat we are interested in and get their position and size.Target detection has always been the most challenging problem in machine vision because of the different appearances,shapes,poses,and interferences of illumination and occlusion during imaging.The traditional target detection time complexity is high,and it has been difficult to break through,so the target detection algorithm based on deep learning has become the mainstream of current target detection and recognition research.The main work in this paper is as follows:Firstly,an image target detection model based on fusion convolutional neural network and regression neural network is proposed.This paper introduces a bidirectional parallel image feature extraction model(BPRNN)based on recurrent neural network,and uses its three-layer structure to fuse itself into the image target detection model SSD based on convolutional network.It adds feature structure information,and intergrates the advantages of CNN and RNN in the image processing.And it also replaces the original ReLU function with a more suitable SeLU activation function.After that,a more complete image target detection model is construct.The advantages of the fusion network in image target detection are proved by experiments.Then,the target detection model embedded the SE structure is proposed.The SE-Conv basic block is designed based on the SENet structure.The SE-CNN network is constructed based on the basic block,and the effectiveness of SE-Conv basic block is proved by experiments of the image classification task.Next,the SE-Conv block is used to reconstruct the SSD detection model and the SE-SSD model is proposed.Furthermore,the bidirectional parallel image feature extraction model proposed in this paper is fused into SE-SSD to construct the SE-BPRNN-SSD target detection model,and a series of experiments are provided to prove the effectiveness of the network.Finally,a prototype system for video image detection based on SE-BPRNN-SSD model is proposed.The static image target detection is migrated tothe real-time detection of the video image.The fusion model designed in this paper is used to detect the static image and video image in the real scene,and the complete demonstration and summary is carried out through the design demonstration system.