Method, device, equipment and storage medium for brain-inspired imaging

Abstract

The present application discloses a method, device, device and storage medium for brain-like imaging, wherein the method includes processing the acquired image, extracting instance-level information; judging the information category of the instance-level information; In this case, the semantic information base is updated through the brain-inspired neural network according to the instance-level information corresponding to the valid information; when the information category includes inferable information, the new information in the image is determined according to the instance-level information corresponding to the inferable information. Through the above-mentioned processing methods, data redundancy can be reduced, bandwidth and computing power can be allocated more efficiently, and reasoning about new information can also be realized, so as to realize efficient perception and understanding of multiple objects in a large scene.

Description

technical field [0001] The present application relates to the field of neural networks, and in particular to a method, device, device and storage medium for brain-inspired imaging. Background technique [0002] In large dynamic scenes, the expression and understanding of natural phenomena requires high perceptual resolution. Although machine imaging in the prior art has reached 100 million-level resolution, it only focuses on the perception of pixel sets. The way to improve the perception resolution, the space-time / hardware complexity is high, the amount of data is large, and it needs huge bandwidth and computing power as support. Moreover, the global understanding of the scene based on local perception is only a feed-forward one-way path, and the overall perception and understanding ability of the imaging system is severely restricted by the performance of local perception, which faces the problems of low robustness and poor adaptability. Contents of the invention [000...

AI Technical Summary

Problems solved by technology

[0002] In large dynamic scenes, the expression and understanding of natural phenomena requires high perceptual resolution. Although the machine imaging in the existing technology has reached a resolution of 100 million, it only focuses on the perception of pixel sets. The way to improve the perception resolution, the space-time / hardware complexity is high, the data volume is large, and a huge bandwidth and computing power are required as support

Moreover, the global understanding of the scene based on local perception is only a feed-forward one-way path, and the overall perception and understanding ability of the imaging system is severely restricted by the performance of local perception, facing the problems of low robustness and poor adaptability

Images