Research On Application Of Convolutional Neural Network In Object Detection Algorithm

Object detection technology is one of the key technologies that the robot visually perceives external information and interacts with the outside world.It is the basis for the robot to realize high-level visual tasks such as scene understanding.This paper discusses and researches the existing object detection algorithms,and focuses on the object detection algorithm based on convolutional neural network.The specific research contents include: the structure and optimization method of convolutional neural network,feature extraction based on convolutional neural network.Improvement of Network,Object Detection Algorithm and Experimental Evaluation.Firstly,the typical structure of convolutional neural network is deeply analyzed and a typical convolutional neural network model is constructed.Based on the convolutional neural network model constructed,the activation function of the convolutional neural network,iterative algorithm,and the regularization scheme of the convolutional neural network are studied.Experiments are performed on the CIFAR-10 data set.Secondly,based on the research of Vgg16 network model,an image classification model with greatly reduced computational and parameter scales was proposed and trained by ImageNet dataset.In this test set,a classification accuracy rate of 72.4% was obtained,and it can be verified that this model can still obtain image classification accuracy equivalent to that of the existing model in the case of a large reduction in calculation amount and parameter scale,in order to improve the real-time performance of the object detection.Foundation.Finally,based on the proposed image classification model,an improved object detection model was proposed based on the SSD object detection model.The PASCAL VOC data set was used for training and evaluation.The results showed that the proposed improved model compared to SSD object detection.The model has greatly improved the real-time performance under the condition that the detection accuracy rate is not much different,which lays a good foundation for the intelligent research of the robot visual perception system.In this paper,the SSD object detection model is improved by reducing the calculation amount and parameter scale of the feature extraction network without reducing the precision.The research content has more important significance for accelerating the intelligentization of the robot.