Calculation method of adaptive variable-scale convolution kernel based on sonar 3D image confidence

Abstract

The invention discloses an adaptive scale-variable convolution kernel calculation method based on sonar three-dimensional image confidence. According to the confidence of the target object in the three-dimensional image acquired by sonar, the scale and number of three-dimensional convolution kernels are calculated, which not only solves the problem of The problem of uneven distribution of feature information of images extracted in space is solved, and the problem caused by the distance of sonar imaging can also be overcome. At the same time, the invention can reduce the amount of calculation and has better effect.

Description

technical field [0001] The invention relates to the field of image information, in particular to an adaptive variable-scale convolution kernel calculation method based on the confidence degree of a sonar three-dimensional image. Background technique [0002] With the introduction and development of deep learning, people often use convolutional neural networks for image processing to achieve target detection, image classification and other purposes. [0003] In the convolutional neural network, it is necessary to set different convolution kernels to extract the features of different levels of the image. In the traditional two-dimensional convolution, the size of each convolution kernel, the number of channels, and the number of convolution kernels are selected through experiments. There is no theoretical support for the optimal value. The same is true for 3D convolution. When processing three-dimensional sonar images, the accuracy of the target object generated in the image...

AI Technical Summary

Problems solved by technology

If the traditional method is used to perform three-dimensional convolution on three-dimensional images generated at different distances, the effect is definitely not satisfactory.

Images