Design Of Edge-end Helmet Recognition System Based On Deep Learning

With the progress of the times and the innovation of science and technology,artificial intelligence and deep learning computing platform has been developed rapidly.By deep learning can be achieved through the construction of multi-layer neural network to learn and understand complex data,so as to improve the performance of the machine on a variety of tasks,in recent years many research scholars have carried out research based on deep learning helmet wearing recognition system,consider using this intelligent helmet wearing recognition system instead of manual detection,but most of them have not considered the actual cost of deep learning computing server,in the actual In the application there are problems such as high system deployment cost and high delay in uploading monitoring image data.In this paper,based on the current status of the helmet wearing recognition system,we propose a solution based on deep learning for the edge-end helmet recognition system,so that it can complete real-time image acquisition at the edge-end,and use the deep learning-based model to carry out helmet wearing recognition on the collected images,and feed the recognition results to the cloud platform,which reduces the system application deployment cost and improves the system detection operation efficiency.This paper proposes an improved Mobile Net-SSD helmet wear recognition algorithm to address the problems of many small proportional targets,obscured targets and complex background environments in existing implementation schemes while ensuring the efficiency of helmet wear recognition.1.The algorithm introduces a mixed domain attention mechanism to enhance the features of multiple dimensions of image information,suppressing non-focused information and focusing on The algorithm improves the recognition success rate of obscured targets by suppressing the influence of complex background environment on helmet wearing recognition.2.By introducing dilated convolution,the low-level feature perceptual field is improved,resulting in less feature loss for small-occupancy targets in images and improving the accuracy of the algorithm network for small target recognition.The algorithm in this paper has better robustness and higher recognition accuracy than the original Mobile Net-SSD algorithm in different recognition scenarios.Finally,based on the design of the helmet wearing recognition algorithm,this paper implements the deployment of the algorithm model on the edge-end device,and completes the construction of the whole system architecture,and realises the data upload function through HTTP.After system testing,it is shown that the edge-end helmet recognition system can basically meet the needs of practical applications in terms of real-time and reliability,and has certain engineering application value,and also provides a practical case for edge computing based on deep learning.