Indoor scene recognition method based on SVM and reinforcement learning

Abstract

The invention discloses an indoor scene recognition method based on SVM and reinforcement learning. The method comprises the following steps: (1) collecting data of the indoor scene and training an SVM classifier Ns capable of classifying the indoor scene; (2) Conducting several indoor scene active recognition experiments. In the experiment, a reinforcement learning neural network fitting the reinforcement learning value function is trained, and the network is named decision network NQ. The decision network NQ judges the classification result by Ns. (3) After the training of decision network NQ, Decision-making network NQ is used to make decision and execute the robot action according to the room laser ranging information currently obtained by the laser ranging sensor. After the action isexecuted, the information of the laser ranging sensor is collected again and inputted into NQ. Thus, the result obtained three times is fused to obtain the final classification result. In this method,the laser ranging information is transformed into the projection information of the scene contour map ring and the SVM is used for scene recognition, which reduces the computational load and improvesthe correct rate of scene recognition.

Description

technical field [0001] The invention relates to an indoor scene recognition method, in particular to an indoor scene recognition method based on SVM and reinforcement learning. Background technique [0002] In recent years, robotic scene recognition has been used in an increasing number of robot localization functions. The robot recognizes the type of environment as soon as possible in an unfamiliar environment, laying the foundation for it to realize various other functions such as path planning and behavior control. In the current application scenario, the laser sensor equipped on the robot can only obtain the distance information of 180° ahead. Due to the limitation of the robot's orientation and the data limitation of the low-cost sensor, the accuracy of scene recognition is low. [0003] There is an invention patent application document. The subject of the patent application is "A Scene Recognition Method Based on Single Hidden Layer Neural Network". The application pu...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide an indoor scene recognition method based on SVM and reinforcement learning. In order to solve the problem that the existing technology can only rely on the richness of collected sample data to improve the recognition accuracy and the workload is huge

Images