Research And Implementation Of Apple Grading Technology Based On Visual Inspection

As the world’s leading producer of apples,China’s apple production and exports are on a yearon-year rise.As an important part of the apple production process to achieve high profitability,apple grading has significant labour-intensive characteristics.The existing apple grading equipment at home and abroad has problems such as low intelligence,low grading accuracy and large equipment size,which are not suitable for the actual grading needs of small and medium sized businesses in China.To solve this problem,this paper starts the research from the hardware construction of apple automatic grading machine,grading algorithm accuracy improvement and system solution testing,proposes the apple grading method based on improved YOLOv5,and designs and develops the apple automatic grading machine,and verifies it through experiments.The main research in this paper contains the following points.(1)The current status of fruit grading research is introduced.The two major categories of current fruit grading algorithms based on machine learning and deep learning are described.The advantages and disadvantages of the two grading methods are compared to provide a research direction for the grading algorithms in this paper.The advantages and disadvantages of existing fruit grading equipment at home and abroad are explained,providing directions for the design of the automatic apple grading platform in this paper.(2)An experimental platform for an automatic apple grading machine is built,and constructed a multi feature apple dataset.Based on a detailed analysis of the requirements,a general scheme design is given and the hardware structure of the automatic apple grader is described in detail.The main introduction is the loading and unloading mechanism,the turning and conveying mechanism,the visual inspection and automatic grading control system and the grading actuator,which provide the hardware platform basis for the subsequent experiments in this paper.The required experimental equipment and materials were introduced,the collection of the red Fuji apple dataset was completed,and the dataset was expanded using mirror flip and other methods to increase the amount of data in the apple model for different grades and to improve the robustness of the apple grading model.Based on national standards and existing grading experience,this paper developed multi-characteristic apple grading criteria and labelled the apple dataset.(3)Developed a deep learning based model for grading comprehensive Apple features.Optimized the Mosaic data enhancement part in YOLOv5.Firstly,A CLAHE-Mosaic data enhancement method is proposed to reduce the problem of accuracy rate degradation caused by random data cropping;Secondly,the YOLOv5 backbone network is optimised.The Mish activation function is used to replace the original YOLOv5 activation function,enabling the flow of apple feature information in the deep network and improving the generalisation ability of the model.The distance intersection ratio loss function(DIo U＿Loss)is used to speed up the rate of border regression and improve the speed of model convergence.In order to refine the model to focus on apple feature information,a channel attention mechanism(SE)was added to the YOLOv5 backbone network to enhance information propagation between features and improve the model’s ability to extract fruit features.The results show that the average detection accuracy of the improved YOLOv5 algorithm is 93.46,which is 4.2% higher than the original algorithm,and the real-time detection speed reaches 58 FPS,with high real-time performance and detection accuracy.(4)The establishment and experimentation of the control system of the automatic apple grader.Firstly,the control system of the automatic apple grader was established,and the control system components were introduced in detail.Secondly,the software design of the apple automatic grading machine was completed.The Python-snap communication process was introduced in detail,and a multi-functional apple automatic grading software based on Py Qt5 was designed to facilitate real-time output of apple grade and location information and guide the grading actuator to complete the grading operation.Finally,experiments on the automatic apple grading machine were conducted to introduce the grading workflow,and apples graded manually were used as experimental samples to achieve efficient grading of apples and to verify the feasibility and accuracy of the apple grading algorithm.