Research On The Work Quality Monitoring System Of Unmanned Transplanter Based On Machine Vision

With the continuous improvement of the level of agricultural intelligent mechanization,research on independent operation of unmanned agricultural machinery is also developing rapidly.Timely and accurate monitoring of the planting condition and working conditions of the rice transplanter,as well as real-time alarm when a fault occurs are important guarantees for the safe and efficient operation of the rice transplanter,which can effectively improve the operation efficiency and reduce safety risks.This paper designs and develops a set of machine vision-based unmanned rice transplanter operation quality monitoring system with the support of the research fund of the key project of Jiangsu Province's key research and development plan(modern agriculture)"Development of unmanned rice transplanter"(BE2015351).The rice transplanter is an important method for the development of modern agricultural automation.Due to the influence of factors such as geographical environment and equipment,the rice transplanter will inevitably have shortages and floating seedlings in the work.The traditional identification of missing seedlings and floating seedlings mainly relies on experience and manual work,and the efficiency is low and the accuracy is not high.Therefore,based on the relevant algorithms of machine learning and deep learning,a set of effective identification of missing seedlings and floating seedlings is designed.surveillance system.The main research contents of this article are as follows:1.Preparation of samples of missing and floating seedlings.Because the paddy field sample does not have a public data set,this article collected the original pictures of the paddy field in the paddy field,and then combined the OpenCV database to label the seedlings,and produced a large number of data sets,which laid the foundation for subsequent training and testing.2.Design and implementation of lacking seedling recognition algorithm based on traditional machine vision.First,the RGB seedling-based color channel algorithm is used to separate the green seedlings and the background in the collected seedling pictures,and then the corrosion expansion algorithm is used to optimize the sample pictures,and then the centroid position of the seedling data image sample is calculated.According to whether the distance between the centroids is Identify the lack of seedlings within a reasonable range.3.Design and Implementation of Recognition Algorithms for Missing and Floating Seedlings Based on Deep Learning.First,extract the characteristics of the seedling samples to establish a sample library,analyze and process the collected seedling image data,and then compare it with the sample library to determine whether there is a lack of seedlings and floating seedlings in the transplanting machine.In this paper,the simulation examples are tested to verify the effectiveness of the algorithm.This paper also makes an in-depth study on the use of YOLOv3 algorithm to identify floating seedlings.It is a representative algorithm in convolutional neural networks,its accuracy in identifying floating seedlings is slightly higher than that of RCNN algorithm.4.Realization of secondary positioning technology for paddy field seedlings.Perform image enhancement processing on the initial seedling pictures,reduce noise in the image,improve accuracy,and then extract the outline of the normal seedlings,complete the fitting of the seedlings by the obtained seedling outline information,and further improve the accuracy of paddy field seedling positioning.5.Experimental analysis of the identification and positioning algorithms for missing and floating seedlings.In this paper,the RGB-based color channel,RCNN and YOLOv3 three algorithms to identify the lack of seedlings and floating seedlings were tested comprehensively.Based on the above experimental conclusions,the correct rate of identifying the lack of seedlings based on the RGB color channel is 82%,the correct rate of identifying the lacking and floating seedlings by the RCNN method is 88%,and the correct rate of identifying the lacking seedlings by YOLOv3 is 97%.The accuracy rate is 90.2%,and the recognition accuracy rate is the highest among the three methods.Among them,the recognition algorithm based on YOLOv3 is more ideal in real-time and accuracy.This article also sets up an alarm device for the operation quality monitoring system of unmanned rice transplanter based on machine vision.When it is detected that there is an abnormal seedling in the paddy field or the operation route of the rice transplanter deviates from the normal track,an alarm will be issued to remind the staff to issue a stop or obstacle avoidance command to the rice transplanter.The operation quality monitoring system of unmanned rice transplanter based on machine vision designed in this paper.It provides theoretical methods and data references for identifying the state of seedlings,which is of great significance and practicalpromotion for the development of agricultural automation.