Small target detection method based on attention mechanism

Abstract

The invention discloses a small target detection method based on an attention mechanism, and the method employs an improved Resnet network as a feature extraction network, and enables the Bottle Net network architecture of the Resnet network to be decomposed into a plurality of uniform branch structures, thereby reducing the amount of hyper-parameters, and obtaining a better effect. Space and channel attention is introduced, information among multiple layers is fused, and an improved FPN is used for multi-scale prediction, so that the model can detect small targets and also can detect medium and large targets; each detection output predicts a conditional probability value for each category, and a prediction result is directly obtained from the picture, so that target information is obtained; and transmitting the feature maps of the three scales to a detection head for joint training. According to the method, the deep neural network is used for detection, different data sets are obtained according to different application scenes for training, various different fields can be used, and high detection accuracy is kept.

Description

technical field [0001] The invention belongs to the technical field of biological feature authentication, and relates to a small target detection method based on an attention mechanism. Background technique [0002] Object detection is also one of the four basic tasks of computer vision, and it has a very broad application prospect. Target detection technology has great application value in both military and civilian fields, such as in important occasions such as airports, railway stations, ports, and UAV ground detection, as well as video surveillance, face recognition, intelligent transportation, etc. There are applications, and have achieved good results, but also provide a technical basis for tasks such as image analysis, understanding and behavior recognition. However, the technology is not perfect, and there are some difficult problems, such as the problem that small targets are difficult to detect. This problem is common in daily life, such as relatively small vehic...

AI Technical Summary

Problems solved by technology

The third is that the size and aspect ratio of the anchor frame in the existing anchor frame-based target detection method are set based on medium and large targets, so that small targets are ignored in the entire learning process, and the receptive field in general target detection It is not very friendly to small targets. The receptive field of small target features mapped back to the original image may be larger than the size of the small target in the original image, resulting in poor detection effect

However, the FPN network simply superimposes the feature map obtained by the backbone network and the feature map obtained by top-down sampling to obtain a new feature map. The spatial information and channel information in the feature map have not been fully utilized.

Images