Research On Real-time Multi-objective Recognition Technology Based On Field Environment

With the continuous development of artificial intelligence technology,using deep learning technology to achieve real-time multi-target detection in the field environment has gradually become an important research direction.In this paper,the current traditional image recognition algorithm in the complex field environment generally has the defects of slow recognition speed and low accuracy,especially in the changeable climate and more occlusion of the field environment.Faster R-CNN＿F is designed.This algorithm optimizes the Faster R-CNN network to improve the detection accuracy and recognition efficiency.The main research contents are as follows:(1)Aiming at the characteristics of low recognition rate and high false alarm rate in foggy days,the improved haze removal algorithm based on dark channel is adopted.The image distortion,unclear target and feature disappearance are improved by down sampling,edge extraction,feature enhancement and contrast enhancement;Aiming at the weakness of small feature extraction ability of RPN,multi-resolution feature maps extracted from different network layers are used to ensure good detection accuracy of targets on different scales;The NMS algorithm is improved to improve the recognition accuracy and speed of occluded targets;The improved algorithm adds part of semantic information,optimizes the position of the target candidate box through weak supervised detection,and eliminates the wrong target.(2)Aiming at the characteristics of low recognition rate and high false alarm rate in foggy days,the improved haze removal algorithm based on dark channel is adopted.The image distortion,unclear target and feature disappearance are improved by down sampling,edge extraction,feature enhancement and contrast enhancement;Aiming at the weakness of small feature extraction ability of RPN,multi-resolution feature maps extracted from different network layers are used to ensure good detection accuracy of targets on different scales;The NMS algorithm is improved to improve the recognition accuracy and speed of occluded targets;The improved algorithm adds part of semantic information,optimizes the position of the target candidate box through weak supervised detection,and eliminates the wrong target.In this thesis,through the experiment of the Faster R-CNN＿F network model,the original Faster R-CNN model obtains the detection speed of 0.083s/frame and the average recognition accuracy of 67.38%,while the Faster R-CNN＿F model obtains the detection speed of 0.043s/frame and the average recognition accuracy of 71.34%,0.04 s/frame and 3.96% respectively.The experimental results show that the improved network model has a good recognition effect for real-time multi-target recognition in the field environment,which shows that the algorithm has a certain practical value.