Research On Object Detection And Recognition Of Small Sample Infrared Image Based On Deep Learning

As a basic problem in the field of computer vision,object detection and recognition has been an active research direction.As the cornerstone of image understanding and computer vision,object detection and recognition has gradually become the basis for solving more complex or higher-level visual tasks.Deep learning has brought significant progress to the problem of target detection and recognition,and has attracted great attention and extensive research in the past five years.However,most of the research on object detection and recognition is focused on natural scene images,and the research and application in infrared images are rare,and the research on object detection and recognition in infrared images has great potential in both civil and military fields significance.In this paper,in view of the current lack of infrared image data and the lack of research results in infrared image target detection and recognition,this paper proposes a method of small sample infrared image target detection and recognition based on deep learning.First,an improved single-scale Retinex image enhancement algorithm is added before the deep convolutional generative adversarial network,and the self-attention module is introduced into the deep convolutional generative adversarial network,making it easier for the network model to extract non-local feature information and key Feature information to generate infrared image data with sufficient quantity and good effect,suitable for infrared image target detection and recognition research.Then add the super-resolution algorithm based on the convolutional neural network before the detection model Faster R-CNN based on the region proposal,so that the network model can more easily extract the key feature information of the picture,and introduce it in the convolutional neural network of Faster R-CNN The residual learning module reduces the loss of feature information during training,adjusts the hyperparameters during training,and optimizes the training effect.The experimental results show that the improved method proposed in this paper is better than the traditional Faster R-CNN for the infrared images used in this paper.The detection accuracy of each category is obviously improved,the mean average precision reaches 84.63% and does not cost the excessive detection rate.