Faster-RCNN target object detection method based on deep reinforcement learning

Abstract

The invention discloses a Faster-RCNN target object detection method based on deep reinforcement learning. The comprises the steps: storing the state of the regional suggested network RPN at each moment by adopting an experience pool of deep reinforcement learning, outputting two actions by adopting a convolution gating circulation unit, selectively executing corresponding actions by adopting a random strategy, and removing redundant detection boxes by adopting a self-defined non-maximum suppression method to obtain a detection box closest to a labeling box; classifying the detection frames byadopting a classification network, and carrying out quadratic regression on the detection frames to realize detection and identification of the target object. By adopting the technical scheme, the target positioning is accurate, and the target detection precision is high.

Description

technical field [0001] The invention belongs to the technical field of computer vision and relates to target detection and recognition technology, in particular to a Faster-RCNN target object detection method based on deep reinforcement learning. Background technique [0002] Object detection is to find out all the objects of interest in the image and determine their position and size, which is one of the core problems in the field of computer vision. Object detection has always been the most challenging problem in the field of machine vision due to the different appearance, shape, and posture of various objects, as well as the interference of factors such as illumination and occlusion during imaging. As deep learning has made great progress in image classification tasks, object detection algorithms based on deep learning have gradually become mainstream. [0003] The target detection method based on deep learning is mainly divided into two types: One-stage and Two-stage. ...

AI Technical Summary

Problems solved by technology

[0010]Deep Q-Network(DQN)(Human-level control through deep reinforcement learning[J].Nature,2015,518(7540):529-533.) is a A method of combining deep learning with reinforcement learning. The present invention uses the experience pool of DQN to save the region proposal network state at each moment, and then adopts a convolutional gated recurrent unit (Convolutional Gated Recurrent Unit, ConvGRU) (Nicolas Ballas, Li Yao, ChrisPal, and Aaron Courville. Delving deeper into convolutional networks for learning video representations. In ICLR, 2016.) output two actions, and use a random strategy to select and execute the corresponding action to achieve more accurate detection and identification, which can solve the existing The method has problems such as low precision, and has high recognition precision

Images