Research And Design Of Unsound Wheat Kernel And Impurity Detection System Based On Deep Learning

As one of the important grain crops in China,grading wheat is an important measure to ensure the safety of wheat production,storage and trading.In this process,the content of unsound wheat kernels and impurities is an important indicator that affects the wheat rating.Therefore,it is of great practical importance to carry out research on the detection technology of unsound wheat kernels and impurities for food safety and grain storage.However,at present,the traditional manual detection methods have the disadvantages of large workload,slow detection speed and strong subjective factors;while most of the existing automatic detection equipment is off-line detection with strict image acquisition conditions,and the recognition algorithms are mainly traditional machine learning algorithms,which require manually designed classifiers,with slow detection speed and low efficiency.Therefore,this study addresses the problems in the existing research on unsound wheat kernel and impurity detection,and investigates the detection method of unsound wheat kernel and impurity with wheat imperfection and organic impurity(wheat hull)as the detection target and deep learning target detection technology,and designs and implements a set of unsound kernel and impurity detection system.The main contents and conclusions of the paper are as follows:(1)In response to the problem that there are few experimental datasets for multi-category unsound wheat kernel and impurity target detection algorithms,an automatic wheat image acquisition device was designed and built in this paper.The data sets of unsound wheat kernel and impurity images were constructed by annotating the sample images.(2)The paper investigates the common image processing and target detection methods,and processes the image dataset by histogram equalization,color space conversion,flip,random masking and other image enhancement methods,so that the detection model can obtain better image features and has stronger generalization ability.(3)The paper analyzes and compares the problems of existing common detection methods,and accordingly proposes a wheat imperfection and impurity detection method based on the improved YOLOX model.The method introduces an attention mechanism in the backbone network to enhance the saliency of the seeds in the image,and uses a depth-separable convolution module to lighten the feature extraction network,and uses a bidirectional weighted feature pyramid structure to improve the YOLOX target detection network.The proposed method was experimentally validated on the dataset built in this paper.The experimental results show that the model has high accuracy and fast detection speed for detection of wheat imperfections and impurities,and it can identify and count unsound wheat kernels of multiple categories simultaneously with good generalization ability.(4)To enhance the practical applicability of the research conducted,a wheat imperfection and impurity detection system is developed.In response to the problem that most of the image acquisition parts of the existing detection devices can only collect static wheat grain images with regular arrangement,a dynamic wheat imperfection and impurity acquisition and detection platform is designed using a drop chute,a vibrator,a conveyor,an industrial camera and an open-hole light source,which can continuously collect wheat grain and impurity images.For the interaction between the user and the detection system,a wheat imperfection and impurity detection interface was developed using Pyqt5,and the proposed detection algorithm was built into the system.The proposed detection algorithm was built into the system.It was proved through experiments that the detection system has high recognition rate and fast detection speed,which has good practical application value.