High-reflection noise removal method based on improved U-Net model

Abstract

The invention discloses a high-reflection noise removal method based on an improved U-Net model, which can improve the running speed and the detection accuracy of a high-reflection area in an image, the improved U-Net model comprises three convolution modules of an up-sampling path and a down-sampling path, each convolution module comprises two convolution layers connected with a linear rectification function, and the convolution layers are connected with the linear rectification function. The pooling layers are linked by a maximum pooling layer; the method comprises the steps of obtaining an original image of a to-be-measured object, preprocessing the original image, training an improved U-Net model to construct the improved U-Net model, adding batch standardization layers, adding a standardization layer behind each convolution layer in a convolution module, abandoning neuron operation, initializing weight by using a Xavier initialization method, and outputting a segmented image, the method comprises the following steps: taking original point cloud data, carrying out registration on the original point cloud data, obtaining registered point cloud data, and carrying out noise reduction, smoothing, simplification and three-dimensional reconstruction on the registered point cloud data.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a high-precision visual detection and removal method of high reflective noise based on the U-Net improved model. Background technique [0002] When inspecting the surface quality of industrial parts, metals and other objects, computer vision technology is one of the most commonly used technical means at present, but the surface of industrial parts, metals, mirrors and other objects tends to have highly reflective areas, through computer vision technology The camera in the camera collects images of the surface of objects such as industrial parts with high reflective characteristics, and then uses image processing technology to process the collected images, the highly reflective areas in the image will seriously affect the image processing effect, which not only increases the difficulty of image processing, And it is easy to lead to inaccurate test results. [0003] It is ...

AI Technical Summary

Problems solved by technology

However, the U-Net network also has two obvious shortcomings: (1) The running convergence speed is slow, there is an over-fitting problem, and the original network characteristics will cause the same feature to be trained multiple times, resulting in waste of resources

(2) Positioning accuracy and obtaining contextual information cannot be achieved at the same time

[0005] Compared with the traditional method, the method based on deep learning still relies on training data. When the training data is insufficient, the output result is prone to be poor, and the processed image has problems such as color distortion and gray value residual, and it is difficult to achieve the accuracy required by the industry. standard

Images