Single-stage target detection method based on convolution region reregistration

Abstract

The invention discloses a single-stage target detection method based on convolution region reregistration. The method comprises the steps: acquiring a training set of detection pictures; standardizingpixels of the training pictures and the test pictures of the training set and zooming the pictures to the same size; establishing a deep convolutional neural network structure, and performing training by using a loss function and the training set to obtain a network model; testing the test picture according to the network model, repositioning the convolution kernel sampling area based on the coordinate regression result of the first stage, judging the specific category of the foreground target area on the basis of the feature pyramid, and performing the coordinate regression of the second stage; and carrying out post-processing on the obtained detection result through a non-maximum suppression method to obtain a final detection result. According to the invention, the receptive field of the convolution kernel during the second-stage detection can be adaptively adjusted according to different sizes of the targets in the picture, and more complete feature information is obtained, so thata better target detection effect is achieved.

Description

technical field [0001] The invention relates to the technical field of deep neural network target detection, in particular to a single-stage target detection method based on convolutional region re-registration. Background technique [0002] With the development of deep learning technology, the target detection technology based on deep learning has attracted more and more attention, and the scope of application has gradually expanded, ranging from the camera of a smartphone to the assembly line of industrial production. The land of martial arts. Traditional target detection technology requires professionals to design corresponding feature extractors and classifiers for specific application scenarios, such as HOG features (abbreviated for Histogram of Oriented Gradient, representing histogram of oriented gradients) and SVM classifiers (abbreviated for Support Vector Machine, stands for support vector machine), etc. However, the application scenarios of traditional target de...

AI Technical Summary

Problems solved by technology

[0004] The disadvantage of the existing technology is that in many practical application scenarios, the requirements for detection accuracy are not very strict, but a sufficiently fast detection speed is required, especially in some real-time detection projects

However, when the scene changes greatly and the size of the target to be detected changes drastically in the image, especially when the target to be detected has different rotation angles, the ordinary single-stage detection method cannot solve these problems well, and it is easy to cause The missed detection and false detection of the target will affect the final detection effect

Images