Real-time target detection method combining convolutional neural network and Transform network

Abstract

The invention provides a real-time target detection method combining a convolutional neural network and a Transform network, and belongs to the field of image processing. Comprising the following steps: S1, inputting image data; s2, the image passes through a convolutional neural backbone network, so that the extracted features have inductive bias characteristics; and S3, designing a neck detection network, performing transition between the detection backbone network and the head network, and providing high-resolution and high-semantic features for the detection head network. S4, designing a detection head network, introducing Transform into the head network, constructing a plurality of remote dependency relationships among the generated local features, and representing target categories and coordinates existing in the image; s5, designing a nonlinear combination method for reducing false negative samples and improving the capturing capability of the detection model on the target; and S6, carrying out detection on the natural data set. Based on the method, better performance is realized on challenging PASCAL VOC 2007, 2012 and MS COCO 2017 data sets, and the method is superior to many more advanced real-time detection methods.

Description

technical field [0001] The invention belongs to the field of image processing, and relates to a real-time target detection method combining a convolutional neural network and a Transformer network. Background technique [0002] Object detection is an attractive and challenging topic in computer vision. Its appeal comes from a wide range of applications such as autonomous driving and robotic navigation, while the challenges come from changing scales, complex shapes, and multiple categories. With the rapid development of Convolutional Neural Network (CNN), the number of object detection models has increased rapidly. Although the models are diverse, they can all be classified into anchor-based methods or anchor-free methods by the depth stack of convolution operations, which are more sensitive to local regions of interest than Multi-Layer Perception (MLP) methods. Fewer parameters are required. However, a striking feature of these methods is that the image features extracted...

AI Technical Summary

Problems solved by technology

[0006] In order to solve the training and prediction deviation introduced by the local features of the convolutional neural network, and the problem of convergence difficulties in the training of the Transformer network in the limited data set, the present invention proposes a method combining the convolutional neural network and the Transformer network to solve the problem. The above problems, so as to realize the real-time accurate detection of objects

Images