Data processing method and device based on automatic lightweight neural network

Abstract

The invention relates to a data processing method and device based on an automatic lightweight neural network. The method comprises the steps of obtaining to-be-processed data; inputting the to-be-processed data into a pre-established automatic lightweight neural network, and obtaining a processing result of the to-be-processed data output by the pre-established automatic lightweight neural network; according to the technical scheme provided by the invention, the automatic lightweight neural network has rich neural network topology connection forms; the method is advantaged in that the searched network is more diversified, through cooperative use of multiple convolution, not only can the network be more lightweight, but also the receptive field of the network is more flexible, and identification, detection and segmentation of objects or scenes with relatively large scale changes in the data processing process are facilitated.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a data processing method and device based on an automatic lightweight neural network. Background technique [0002] The construction of large-scale deep neural networks usually requires strong expert knowledge, and generally consumes a lot of time and energy for researchers, especially in the field of remote sensing image processing. At the same time, these artificially designed networks usually have high computational complexity and large memory overhead, which also poses great challenges to deployment on edge computing devices. Neural Architecture Search (NAS) is an automated design method specifically for deep neural networks. This method can effectively reduce the degree of manual participation and automate the construction of deep neural networks from the perspective of machines. [0003] The current NAS research mainly uses reinforcement learning (Reinforcement Learning, RL)...

AI Technical Summary

Problems solved by technology

[0004] However, the network topology connection form derived by RL-based NAS is relatively simple, and the computational complexity of the model is high, which is not suitable for large-scale, multi-scale remote sensing images

At the same time, the optimization cost of RL is also very high. Usually, this method requires parallel computing on hundreds of computer graphics processing units (GPUs) for about four weeks, and there are also problems such as instability in the search process.

Images