Improved few-sample target detection method based on Faster RCNN

Abstract

The invention provides an improved few-sample target detection method based on a Faster RCNN (Regional Convolutional Neural Network). On the basis of a traditional target detection framework Faster RCNN, deep optimization and improvement are carried out on the Faster RCNN in combination with a CBAM attention module, a global-local relation detector and a classifier based on cosine Softmax loss, so that the Faster RCNN is beneficial to few-sample target detection. According to the invention, the CBAM attention module is combined with the RPN network in the Faster RCNN to form the CBAM-Attention-RPN network based on the attention mechanism, thereby facilitating the generation of candidate boxes of a specific category, and improving the precision of a subsequent network. According to the method, a global-local relation detector is provided, and the support image features and the query image features are subjected to feature matching by using the global relation and the local relation, so that the candidate box more related to the target category can be obtained. According to the method, a classifier based on cosine Softmax loss is provided as a classification branch, so that intra-class variance can be reduced, and the detection precision of a new class can be improved.

Description

technical field [0001] The invention relates to the fields of deep learning target detection and few-sample learning, in particular to an improved few-sample target detection method based on FasterRCNN. Background technique [0002] As an important computer vision task, object detection aims to find out the objects of interest in the image, determine their location and category, and is the basis of many other computer vision tasks. In recent years, with the emergence of powerful computing equipment, large-scale data sets, and advanced models and algorithms, object detection based on deep learning technology has developed rapidly and gradually replaced traditional detection methods. Now, object detection has been widely used in many practical applications, such as autonomous driving, robot vision, video surveillance, etc. Although many mature target detection algorithms have been practically applied, their drawbacks have also begun to emerge. One of the major problems is tha...

AI Technical Summary

Problems solved by technology

Although many mature target detection algorithms have been applied in practice, their disadvantages have also begun to emerge. One of the bigger problems is that the application of most mature algorithms requires the help of large-scale labeled data, and in the vast majority of In most practical application scenarios, collecting labeled data that meets the requirements is a very costly work of financial and material resources, which not only leads to problems such as single application scenarios and single coverage tasks, but also greatly limits the application and promotion of target detection methods. , so how to use less labeled data to learn a model with a certain generalization ability has become an urgent research problem

Images