Research And Application Of Object Detection And Segmentation Based On Deep Learning

In recent years,artificial intelligence has become one of the hot topics in people's daily life.With the development and progress of science and technology,artificial intelligence is applied to many fields such as scientific research and engineering.As an important branch of artificial intelligence,deep learning is a hierarchical operation mode.Its development is mainly inspired by the new discoveries of biology and neurology in the field of animal and human visual nerves.Deep learning mainly extracts high-level abstract feature representations from the original input data,and these representations have good robustness and generalization ability,overcoming some of the difficult problems that were considered in artificial intelligence in the past.With the growth of training data sets and the continuous development of hardware facilities,target detection and image semantic segmentation based on convolutional neural network solve the bottleneck of traditional target detection methods,and become the mainstream algorithm for detection and segmentation.Compared to traditional target detection algorithms,feature extraction and classification decisions in convolutional neural networks are performed simultaneously to achieve end-to-end learning.With the increase in the number of layers,it is possible to better handle more complex scenarios.Therefore,how to study effectively It is of great value to use convolutional neural network for target detection and image segmentation.In this paper,the traditional algorithm and target detection based on convolutional neural network are deeply studied.Through the design and optimization of the network structure,it is applied to the detection and segmentation of the strip defects.It has achieved a certain degree of practicality.The main research contents and results of this paper are as follows:(1)The strip defect detection model based on convolutional neural network was realized.In this paper,through the traditional feature extraction algorithm,classifier selection criteria and convolutional neural network for the detailed learning of feature extraction,and through the experimental comparison of the full connected layer and full convolution,the optimization of the network model is realized,and The training of the network model was carried out in the equipment provided by the laboratory and successfully applied to the detection of strip defects in the strip to achieve real-time inspection.(2)The network model of image semantic segmentation based on deep learning is realized.We know that the general detection method achieves the positioning through the recognition of the target.However,in actual applications,we need to know the target's structure.Semantic image semantic segmentation based on deep learning can effectively classify and detect image pixels.Segmentation of the instance of the target can be achieved.However,the current image semantic segmentation algorithm has relatively limited application scenarios.In this paper,the SegNet algorithm is improved.Residual link memory multi-scale feature fusion is used to reduce feature redundancy and reduce feature dimension.At the same time,the VGGNet network is replaced by AlexNet network,which accelerates the convergence speed of the network and improves the speed of the algorithm.And applied to the strip defect detection,it can achieve the segmentation of defects,and the boundaries are obvious.(3)Target detection and segmentation system based on deep learning.This article adopts open source system Linux,open source deep learning framework Caffe and open source programming language library Python to realize the design of the system,which makes the work of this article have certain engineering and practical value.