Research On Maize Seed Quality Inspection Based On Machine Vision

Corn is one of the main grain varieties produced in China,and high-quality corn seeds have an important impact on corn yield.Quality problems such as deterioration and mold in the storage and transportation of corn seeds are inevitable,so in the circulation process of corn seeds,efficient seed purity identification instruments and equipment are urgently needed.In this thesis,machine vision technology is used to study the quality inspection of maize seeds,which lays a foundation for the realization of automatic seed selection machinery.The main research contents include:Firstly,the collected maize seed images are preprocessed,the grayscale image enhancement method is used to improve the image quality,and the median filtering method is used to smooth the noise reduction process of the maize seed images to enhance the image quality.The enhanced image is binarized,and the Otsu(maximum between-class variance)method is used to denoise small particles by combining the open and closed operations in image morphological processing,so as to obtain a binary image with a clearer outline.Then,multi-object contour tracking is used to extract the outline of each corn seed,and the marked seed outline is used to extract the features of corn seeds,and the threshold is set according to the area feature value and color feature value of each seed extracted,so as to detect and mark the seed image containing damage according to the area,and the seed image containing deterioration according to the color feature for quality detection.Secondly,the neural network is used to detect and classify the images of corn seeds containing spoilage,and the captured images are enhanced according to the image preprocessing method.Then,contour tracing is used to extract each corn seed to make a small data set,according to the advantages and disadvantages,corn seeds are divided into three categories,12 color features of the dataset seeds are extracted,the dataset is divided into training set and test set,and after normalizing the feature values of the extracted data set,the traditional BP neural network is used to identify the advantages and disadvantages of the seeds,with an accuracy of 87.5%.In order to improve the accuracy,the sparrow search algorithm SSA is used to find the optimal weight threshold to optimize the BP neural network.The optimized neural network was used for training and testing,and when the initial population size was 30 and the maximum evolutionary algebra was 50,the recognition accuracy of the optimized SSABP network model was 98.44%.The accuracy rate has been greatly improved.Finally,the YOLOv5 model is used to detect the quality of maize seed images by using the object detection method.Firstly,2000 corn seed datasets were crawled,700 pictures were selected as annotation data by cleaning,and the training set,verification set and test set were divided at the same time,and the YOLOv5 s model was finally selected by comparing the performance of each network model of YOLOv5,and then the CA attention mechanism was added to the backbone network of YOLOv5 s to train the corn seed image training set,and the trained YOLOv5s-CA model corn seed image test set was used for detection,and the detection effect was good.Finally,a visual interface(GUI)is established to feed images and videos containing corn seeds into the detection system.