Potential target area detection method based on convolutional neural network and conditional random field

Abstract

The present invention discloses a potential target area detection method based on a convolutional neural network and a conditional random field. The method comprises the following steps of: 1) performing color space and geometric space conversion of the scene images, taking the images as input of the neural network after de-mean processing; 2) constructing a structural network with fusion of a convolutional neural network and a conditional random field, and generating three-way output on the final convolutional layer of the convolutional network, wherein the first-way output is coordinates ofM candidate frames, the second-way output is the binary tag of each candidate frame, the third-way output is similar features of each candidate frame, the three-way output form input nodes of a total-connection conditional random field to obtain the posterior probability of each candidate frame through global optimization; and 3) performing non-maximum inhibition according to the posterior probability to obtain Top K final results. The potential target area detection method based on a convolutional neural network and a conditional random field can obtain a target area detection result with higher location precision and better robustness through optimization of multi-image input and the conditional random field.

Description

technical field [0001] The invention relates to the technical fields of deep learning and computer vision, in particular to a detection method for a potential target area based on a convolutional neural network and a conditional random field. Background technique [0002] With the rapid improvement of computer computing power, fields such as computer vision, artificial intelligence, and machine perception are also developing rapidly. As one of the basic research problems of image object detection, potential target region detection has also been greatly developed. It is to find and locate those windows that are most likely to contain targets in an image, and then use these windows to speed up the target detection algorithm of the image. [0003] Although there are a variety of potential object region detection methods, including methods based on traditional machine learning and deep learning based on convolutional neural networks. However, statistical experiments have shown...

AI Technical Summary

Problems solved by technology

Traditional methods generally only rely on RGB images for learning, so they are easily disturbed by light changes, object color changes, and noisy backgrounds, and are not robust enough in practical applications.

Images