Image recognition method based on cascaded downsampling convolutional neural network

Abstract

The invention discloses an image recognition method based on a cascaded downsampling convolutional neural network. The method comprises the following steps: constructing an image recognition model based on the cascaded downsampling convolutional neural network, wherein the convolutional neural network based on cascade downsampling comprises a main body network and two independent prediction networks, the main body network comprises a plurality of convolution layers and a plurality of cascaded down-sampling blocks, each independent prediction network comprises a category prediction sub-networkand a bounding box prediction sub-network; performing non-maximum suppression on the prediction result of the independent prediction network to obtain an image recognition result; and training the constructed image recognition model, and performing image recognition through the trained image recognition model. The method can achieve the real-time and accurate recognition of the image.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to an image recognition method based on a cascaded downsampling convolutional neural network. Background technique [0002] With the continuous development of computer vision, the fields of industrial inspection, traffic counting and automatic driving have also developed. The most important technology in these fields is image recognition, that is, real-time detection of objects in images. Most current image recognition methods are based on deep convolutional neural networks, and use sliding windows to extract features from images. The earliest success of deep learning was Yann LeCun's successful application of deep learning to handwritten digit recognition. The image recognition method based on deep neural network has achieved good results on various data sets (for example, ImageNet, Pascal VOC), but the detection accuracy on the MSCOCO data set is still low. Scenes that...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide an image recognition method based on cascaded downsampling convolutional neural network, to solve the technical problems in the prior art, and to be able to accurately recognize images in real time

Images