Use Of Deep Learning And Fuzzy Decision Making In Automated Sorting Of Apple

The supply of good-quality agricultural products is a globally significant objective.Fruit is one of the most important agricultural products in the world.Among them,apple is one of the most productive and widely distributed fruits.The quality level of apples in the market often do not meet our requirements.The sorting process of apple has a significant impact on the quality improvement and added value of the marketed product.The traditional method of fruit grading relies on human perception,in which the quality of fruit has to be judged by visual inspection and experience.This process is subjective,non-repeatable,error-prone,and slow.In recent years,research has been conducted on such areas as sensing of information,classification methods,and dynamic performance.The shortcomings of the detection and grading technologies will seriously affect the value of the product and may also lead to non-compliance with export standards and lack of competitiveness in the international market.In this thesis,advanced technologies of sensing,signal processing,feature extraction,and decision making are proposed and developed to provide significant improvements in speed,accuracy,repeatability of the grading of fruits,through automation.According to the requirements of the investigated subject,an experimental platform for apple sorting is designed and developed,which consists of a transmission system,a sensing system,a control system,and actuators.Considering the dynamic characteristics of the sensor and the integrity of information,machine vision and structured light laser detection methods are selected for non-destructive monitoring of apples.In order not to damage apples during sorting,high pressure gas injection is used to sort apples.Also,the experimental platform is used for image acquisition.In order to facilitate feature extraction and classification of collected images,a series of image preprocessing methods is designed,including image enhancement,gray scale,binarization and edge extraction.The image acquisition system is used to acquire part of the image,and image preprocessing simulation experiments are carried out in MATLAB to verify the effectiveness of the method.For the preprocessed images of apple,a feature extraction method is proposed.In order to obtain the color,texture,shape,and three-dimensional information related to the quality of apples,a feature extraction method based on the color moment,wavelet transform,and laser line extraction is designed.Using these methods,feature extraction of the pre-processed images is carried out in MATLAB.This procedure also verifies the effectiveness of the developed algorithm and obtains the feature values that can fully determine the quality of apples.In order to carry out apple grading,classification methods based on volume set neural network and fuzzy neural network(FNN)are designed,and their basic principles and structures are introduced.By combining fuzzy logic with neural networks,the advantages of the two are further enhanced.In order to reduce the inputs of FNN and achieve a degree of sensor fusion,a method of sensor fusion and data dimension reduction based on 2DPCA is designed.In order to verify the validity of the developed apple sorting method,a simulation experiment of apple sorting is designed.1000 Apple images and corresponding laser images are obtained by using the experimental platform.The method of rotation and translation is used to expand the data to 15000.Classification experiment based on convolutional neural network(CNN)is carried out in Keras,and experiment based on 2DPCA and FNN is carried out in MATLAB.At the same time,1DPCA is used as a comparison.The experimental results show that CNN has obvious benefits in apple color image classification,while 2DPCA and FNN have more advantages in the context of sensor fusion and classification.