Object Recognition Method Based on Dual CNN Networks with ROI Pooling

Abstract

The invention discloses a target recognition method based on a double CNN network with RoI pooling, the steps include: S1. The image to be recognized is obtained through a first-level CNN to obtain a target area window, and through a second-level CNN with a RoI pooling layer Obtain the global feature map of the image to be recognized; S2. Input the obtained global feature map and the target area window into the RoI pooling layer of the second-level CNN, and extract the feature vector of the specified dimension of the target area window; S3. Step The feature vector extracted by S2 trains the classifier, detects the target in the image to be recognized according to the trained classifier, and outputs the recognition result. The invention has the advantages of simple realization method, high target recognition efficiency and high recognition precision, easy adjustment of network parameters and the like.

Description

technical field [0001] The present invention relates to the technical field of digital image processing, in particular to a target recognition method based on a dual CNN (Convolution Neural Network, Convolutional Neural Network) network with RoI (Region of Interest, Region of Interest) pooling. Background technique [0002] The application of various intelligent unmanned systems such as drones and robots is becoming more and more extensive, and the requirements for the speed and accuracy of target detection and recognition of the vision system are also getting higher and higher. Commonly used target recognition methods are based on template matching, based on grammatical structure analysis, based on neural networks, and purely based on traditional statistical methods. The more commonly used methods in image target recognition are the target recognition methods based on convolutional neural networks (CNN). The CNN-based target recognition method is to use the convolution of t...

AI Technical Summary

Problems solved by technology

However, the Fast-RCNN structure only improves the spp structure of SPP-NET, but does not distinguish between the target area and the background area, so it still takes a long time;

[0007] (4) The double-CNN method is to use two-level CNN for image target recognition, in which the first-level CNN has only 5 layers of convolutional layers, which are used to obtain the area window where the target is located, and exclude the background area to reduce the total number of windows. These target area windows are scaled to a fixed size, and then the second-level CNN is used to extract features one by one, and finally the classifier is used to classify and identify; although the double-CNN structure separates the windows of the target area and the background area, only retaining The window of the target area can reduce the processing time, but when extracting features, the feature is extracted by convolution window by window, and in order to ensure that the input of the fully connected layer has a fixed dimension, each target area window needs to be scaled to a certain extent processing, so the recognition speed and accuracy are still not high

Images