A multi-scale target detection method fusing context information

Abstract

The invention discloses a multi-scale target detection method fusing context information, and the method comprises the steps of extracting the characteristics of an input image through employing a deep residual convolutional neural network, and obtaining a candidate box set which corresponds to the input image and is used for target detection through employing an RPN network and an improved non-maximum suppression method; for each candidate box, extracting to obtain convolutional features output by the deep residual convolutional neural network, and extracting the convolutional features outputted by the last convolutional layer of the deep residual convolutional neural network in four directions of upper, lower, left and right twice by adopting an LSTM method to obtain context feature information; performing regularization and splicing operation on the context information and the convolutional features to obtain multi-scale features fused with the context information; converting the multi-scale features into high-dimensional feature vectors by using a full connection layer, and carrying out target classification and border position detection by using a classification layer and a regression layer. The method has the advantages of being high in precision, good in robustness and strong in adaptability for target detection.

Description

technical field [0001] The invention belongs to the technical field of deep learning and computer vision, and in particular relates to a multi-scale target detection method for fusing context information. Background technique [0002] Object detection is an important branch in the field of computer vision. Target detection is widely used. For example, in image recognition, face detection, and artificial intelligence models, target detection technology is used to quickly and accurately identify targets from pictures. The traditional target detection technology uses the sliding window method to intercept image fragments of the same scale, and then extracts features from these fragments, then classifies and regresses image features, and finally obtains them through the non-maximum suppression method (Non-Maximum Suppression, NMS). The position coordinates of the rectangle. This traditional target detection method has relatively low accuracy due to manual feature extraction. ...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem that the time scale of the target detection technology in the prior art is large and the context information cannot be integrated into the final use features of the target detection, the present invention proposes a multi-scale target detection method that fuses the context information; this method can fuse the context information In the final classification feature, the cost of multi-scale feature fusion can be reduced at the same time, which can not only improve the detection accuracy of small-scale targets, but also accurately detect targets under complex backgrounds such as target occlusion. The specific technical solutions are as follows:

Images