Research On Object Recognition Algorithm Based On R-CNN

As an important research direction in the field of computer vision,object recognition is the key technology of computer vision that makes computers have the visual ability similar to the human eye.With the development of computer vision technology,object recognition technology not only plays a crucial role in various application scenarios such as security and traffic control,but also plays a key role in giving computer the ability to understand image semantics.Since the emergence of deep neural network technology in 2006,deep learning has gradually become the main method of computer vision.Convolution neural network is the most important part of deep learning method.It has unlimited development potential and excellent performance,and has repeatedly brought a revolutionary breakthrough in the field of object recognition.Therefore,convolution neural network has become an important hotspot in the field of object recognition.In this paper,through the research on object recognition algorithm of regional convolution neural network based on convolution neural network,we can describe and improve a regional convolution neural network model based on YOLO algorithm.Through the improvement,the performance of YOLO algorithm has been significantly improved,the specific work is as follows:1.By modifying the loss function of YOLO algorithm,large objects and small objects can be more fairly optimized in the algorithm,which avoids the situation that large objects are better optimized than small objects.Compared with the original loss function of YOLO algorithm,the improved loss function is more flexible.2.Aiming at the problem of YOLO algorithm in recognizing the detail features of objects,the characteristics of network extraction are further refined by adding a passthrough layer to the YOLO network layer and using 1*1 convolution kernel to expand the network,which further enhances the YOLO algorithm's ability to recognize detail features and identify details between similar object categories.3.In order to further improve the training rate of YOLO algorithm,this paper improves the network training speed of YOLO algorithm by adding BN network layer behind all the convolution networks in the original network of YOLO algorithm.At the same time,by adding the BN network layer,we can reduce the occurrence of network over-fitting.Finally,through pre-training the improved network by using the Pascal VOC 2007 and Pascal VOC 2012 and training the network to improve the recognition of the network on the database by using the author-calibrated database,the network has been experimented by using the database.Through the analysis of the results of the experimental data,the improved YOLO algorithm network has been significantly improved in recognition ability and recognition rate.