Neural network design method for memory calculation and judging whether someone is in room or not

Abstract

The invention discloses a neural network design method for memory calculation and judging whether there is a person indoors, and the method comprises the steps: an ultra-lightweight neural network model is established, i.e., a UL-CNN model; a UL-CNN model is built according to the rules that a chip based on a memory computing architecture is insensitive to high level and the memory is limited within the range of 2K * 2K; a convolution operation method and a full-connection matrix operation method are adjusted by using fixed-point processing, so that the UL-CNN model is applied to a Conv-Flashchip based on a memory computing architecture for reasoning, and finally whether someone is indoors or not is detected. Compared with other mainstream CNNs, when the problem that someone is in the room and nobody is in the room is solved, the method has the advantages that the parameter amount and the calculated amount are reduced, the calculation time is shorter, but the accuracy is basically unchanged.

Description

technical field [0001] The invention belongs to the field of convolutional neural network design, and in particular relates to a neural network design method for memory calculation and whether there are people in the room. Background technique [0002] Convolutional Neural Networks (CNN) is a type of feed-forward neural network that includes convolution calculations and has a deep structure. It has the ability to learn representations and can classify input information according to its hierarchical structure. At present, the HOG (Histogram of Oriented Gradient, HOG) algorithm is mainly used for indoor human detection. Compared with the HOG algorithm, the CNN-based indoor human detection algorithm has the advantages of less calculation, higher real-time performance, less environmental impact on it, and better performance in image processing. At present, the development trend of the mainstream CNN model is to have a deeper level and a higher number of parameters, which are ma...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a neural network design method for memory computing and whether there are people in the room, so as to overcome the defects of the existing detection technology. And the calculation amount is reduced, the calculation time is shorter, but the accuracy rate is basically the same

Images