Study On Defect Detection Technology Of Seedling Tray Based On Machine Vision

Rice is one of the main food crops in China.In order to increase agricultural income,enhance agricultural comprehensive productivity and promote agricultural modernization,one of the important measures is to improve the level of rice planting mechanization.The process of rice planting should mainly adopt mechanized seedling raising and transplanting technology,while the production line of transplanting seedlings is mainly used to complete.As the carrier of rice seedling raising,seedling tray is prone to defects such as lack of blocks,warping and cracks,which seriously affects the efficiency and quality of rice seedling raising production line.At present,in order to select the unqualified seedling trays,manual selection method is widely used,which has a high labor intensity.Therefore,there is an urgent requirement to develop a scientific and reasonable detection system to automatically identify the quality of seedling tray,so as to provide theoretical support and technical support for the automation and intelligent operation of seedling raising.The lack and warping of seedling tray will make the seedling tray working on the automatic precision seedling production line unable to pass through the automatic feeding tray and automatic folding tray mechanism on the production line smoothly,and the seeds and soil in the seedling tray will be sprayed because of cracks,which not only causes unnecessary waste,but also affects the quality of seedling production line.In this paper,the research status of machine vision detection technology used in defect detection at home and abroad is analyzed and summarized.Combined with the defect situation of seedling tray,machine vision method is used to analyze different color models,smoothing methods,image segmentation methods,morphological processing methods,and by extracting HOG features,machine learning model SVM is used to realize different seedling trays.Defect detection.The main research contents and conclusions are as follows:(1)An experimental platform of seedling tray defect detection system was built,and image samples of seedling tray were collected under different conditions.Based on the analysis of the requirements of the existing factory seedling-raising production line,a telescopic dual-camera detection and seedling tray conveying platform was designed;different seedling tray defect information was collected according to the situation of seedling tray defect;when the seedling tray conveying platform was still,different image samples of seedling tray were collected and stored in the computer for the training of machine learning model;when the seedling tray conveying platform was moving,the seedling tray defect information was collected When working in the state,the trained machine learning model is used to test the newly photographed seedling tray image and the actual detection.The experimental platform integrates image acquisition and testing,which provides a basis for the research of seedling tray defect detection technology.(2)According to the defect characteristics of seedling tray,different color models,smoothing methods,image segmentation methods and morphological processing methods were studied.By extracting HOG features,machine learning model SVM was used to detect different defects of seedling tray.By comparing three different color model spaces of seedling tray image,the HSV color model which is advantageous to image segmentation is selected,and the S component is selected for gray processing.By comparing three common smoothing methods,the Gaussian filtering method which is beneficial to noise removal of seedling tray image is determined for smoothing.By comparing three different image segmentation methods,the region growing method is determined to be beneficial to the image segmentation of seedling tray.After the segmented image is processed by morphological closed operation,the matrix is multiplied with the filtered gray image to remove part of the background,and then the HOG features of the image are extracted for the training of machine learning model SVM.The trained model can be used to test the different defects of seedling tray effectively.(3)The study on the defect detection of seedling tray was carried out systematically,and the defect detection of seedling tray was realized.Based on the experimental conditions and purposes,the seedling trays with three defects,i.e.block missing,warping and cracking,were tested.The test results of machine learning model SVM are as follows: the average time of identifying a seedling tray image is 0.633 s.It is also found that the correct recognition rate will increase with the increase of training samples,but the running time of the training part will also increase.Under the condition of the samples collected in this paper,the correct recognition rate is 93.92%.