Rapid visual positioning method for industrial robot

Abstract

The invention discloses a rapid visual positioning method for an industrial robot, and belongs to the technical field of intelligent positioning. Comprising the following steps: acquiring an article data set needing to be positioned, and constructing a convolutional neural network; based on the comprehensive loss function, training the constructed convolutional neural network by using the data set until the comprehensive loss function converges, and obtaining a trained convolutional neural network; fitting an accuracy curve according to the plurality of network fitting accuracy degrees; according to the slope of the accuracy curve, the optimal number of convolution groups containing useless features is obtained; pruning and removing the convolution kernel corresponding to the optimal number of convolution groups containing useless features to obtain a small-parameter convolutional neural network; according to the method, the visual positioning speed of the industrial robot is improved through the small-parameter network obtained by pruning the convolution kernel network parameters corresponding to the convolution layers containing useless features.

Description

technical field [0001] The invention relates to the technical field of intelligent positioning, in particular to a fast visual positioning method for an industrial robot. Background technique [0002] With the development of social economy and the advancement of industrialization, industrial robots are used more and more widely in production. Before performing corresponding operations, industrial robots need to identify and locate the items to be processed. The production cycle is fast on the production line, and the recognition and positioning speed of the robot is required to meet the demand. In particular, robots engaged in handling and assembly work can identify and locate the objects that need to be processed. However, traditional industrial robots often use convolutional neural networks to complete identification and positioning processing. However, due to the large amount of parameters of convolutional neural networks, the efficiency of identification and positionin...

AI Technical Summary

Problems solved by technology

However, traditional industrial robots often use convolutional neural networks to complete identification and positioning processing. However, due to the large number of parameters of convolutional neural networks, the efficiency of identification and positioning using convolutional neural networks cannot meet real-time requirements.

[0003] When using convolutional neural network for visual positioning, there will be some image features extracted, but some image features are less helpful for object recognition and positioning (the existence of this feature will not improve the recognition accuracy a lot, and will also increase a lot of network parameters), It may even reduce the recognition accuracy, such as some noise features in the image, the existence of these features will reduce the recognition accuracy; therefore, the present invention provides a method for adjusting the parameters of the convolutional neural network to reduce the amount of parameters in the neural network, reducing The computing load of the network, and achieve the purpose of improving the speed of network recognition

Images