Research On Object Detection Algorithms Based On CNN For Dairy Goats' Surveillance Video

Object detection algorithms are the basis of intelligent surveillance for farms.Aming at the problem that detection results of those object detection algorithms based on traditional features such as YUV,RGB,LBSP,and texture are unstable,an object detection algorithm combined with CNN and traditional processing technology for dairy goats' surveillance video is studied.This article selected surveillance videos from the dairy goat farm of the Animal Husbandry Teaching Experimental Base of Northwest Agriculture and Forestry University as the research object.By studying a background subtraction method,an image segmentation algorithm and the Faster R-CNN algorithm,a high accuracy object detection algorithm was constructed.Finally,the accuracy of the algorithm was verified by experiments.The main research contents and conclusions of this paper are as follows:(1)A background subtraction method based on self-adaptable parameters.The acquisition and preprocessing of dairy goat videos is the first step of surveillance video processing.In order to reduce the noise existing in the videos and improve efficiency,the preprocessing of the dariy goats' surveillance videos using the image pyramid is studied.Due to the presence of camera jitter,illumination changes,and background disturbances in outdoor scenes,the determination of optimal parameters is difficult.Aiming at the problem that manual adjustment of optimal parameters is labor-consuming and time-costing,this paper proposes a background subtraction method based on self-adaptable parameters.Experimental results show that the proposed method has high accuracy and recall rate.(2)An image segmentation algorithm based on One-Cut.In this paper,the background subtraction results are optimized by the image segmentation algorithm called One-Cut,and the image segmentation algorithm has been optimized by a method named histogram acceleration technology.Because the surveillance video has a large amount of frames per second and the resolution of frames is high,the improvement of the segmentation algorithm's efficiency is very important.In order to solve the problem of low efficiency of One-Cut,this paper combined the histogram acceleration technology,improved the energy function,and optimized the network graph structure.The optimized network graph has lower complexity,and the efficient of our video frame segmentation is higher.Experimental results show that the improved segmentation algorithm has higher subjective and objective evaluation indexes.(3)An object detection algorithm for dairy goats' surveillance video based on Faster R-CNN.The generalization ability of convolutional neural networks has farly exceeded traditional features'.Among the object detection algorithms based on convolutional neural networks,Faster R-CNN is one of the best algorithms.This article briefly introduces the object detection algorithm for dairy goats' surveillance video based on Faster R-CNN.On the dairy goat data set,this paper analyzed three algorithms based on R-CNN,Fast R-CNN and Faster R-CNN respectively.The experimental results show that the algorithm based on Faster R-CNN has higher average precision and efficiency,and the adjustment of parameters has a significant impact on the average precision.(4)The optimization of the object detection algorithm based on Faster R-CNN for dairy goats' surveillance video.Firstly,the hyper-parameters of the convolutional neural network were determined according to the hardware resources.Secondly,Aiming at the problem of false negative result caused by changes in appearance,a spatiotemporal region proposal algorithm combined with the segmentation result of the background subtraction method was proposed.Thirdly,in order to solve the false alarm problem existing in the traditional non-maximal suppression algorithm,a multi-feature-based non-maximal suppression algorithm was proposed.Finally,the VOC data set was introduced to reinforce dataset and the minimum risk loss function was constructed to increase the average precision of the dairy goat category.