Study On Target Detection Technology Applied On Precision Agriculture Based On Machine Vision

Target detection based on machine vision has integreted electronic technology, sensor technology, computer technology, intelligent control technology and so on, which plays a very important role for the realiztion of algricultural machinery automation. Fruit detection is a very challenging problem when developing a blueberry or citrus picking robot. Huanglongbing (HLB) is a great threat to citrus trees during their growth. The identification of HLB is very important for the citrus tree grower. Based on previous research results, this thesis conducted study on machine vision based target detection technology applied on yield mapping of blueberry, green citrus, and HLB detection research. In general, the major works and contribution of this thesis are as follows.(1)An algorithm termed ’color component analysis based detection (CCAD)’ method, was developed. This newly developed ’CCAD’ method for blueberry was proved to be efficient for separating fruit from background and identifying blueberry fruit of different growth stages using natural outdoor color images. Color components in several important color model were analyzed, and three components including R, B, and H were selected to do fruit identification. Cross validation was conducted using not only the traditional classifiers such as K-nearest neighbor (KNN) and naive Bayesian classification (NBC), but another newly introduced ’supervised K-means clustering classifier (SK-means)’. KNN classifier yielded the highest classification accuracy using the prebuilt pixel dataset.(2)This thesis further studied the green fruit detection using RGB images. A fast normalised cross correlation (FNCC) based machine vision algorithm was proposed in this study to develop a method for detecting and counting immature green citrus fruit using outdoor colour images toward the development of an early yield mapping system.Firstly, the background was removed as much as possible based on color component based analysis. Then the potential fruit positions were identified using the proposed fast normalised cross correlation (FNCC) based method. Finally, the number of fruit was determined after combining colour, texture, and shape feature analysis. For a validation dataset of59images,130green fruits were identifies,24fruits were misses,25fruits were false positives. The identification accuracy was84.4%.(3) In this study, a novel method termed ’extended spectral angle mapping (ESAM)’ was proposed to detect HLB.This method extended the traditional spectral angle mapping method by combining endmember extraction method, red edge position technique, and several other techniques. Firstly, the spectral differences between healthy and HLB infected canopies from ground measurement and hyperspectral image was analyzed. Then the performance of the proposed ’ESAM’ method and two other commonly used methods:K-means, and Mahalanobis distance (MahaDist) was evaluated and compared, and it was shown ESAM performed best. A fairly high detection accuracy of82.6%was achieved in the calibration set, and86.3%in the validation set was achieved using the proposed ESAM method. The study also explored the feasibility of using MS image for HLB identification. However, it didn’t achieve as good result as HS image, which indicated that HS image was a better choice for HLB detection.(4) To solve citrus greening disease (Huanglongbing, or HLB) detection problem using airborne HS image, an efficient dimension reduction method needs to be applied. In this study four dimension reduction methods including principal component analysis (PCA), maximum noise fraction (MNF) transformation, forward feature selection algorithm (FFSA) and Kullback-Leibler divergence (KLD) based method, were applied on the obtained HS image. The selected bands or components were used for the following pixel based or tree based classification. KLD showed the highest pixel beased HLB detection accuracy of63.3%using validation pixel dataset, which indicated there was much room for study and improvement of how to use dimension reduction method efficiently. MNF and KLD gave the same tree based HLB detection accuracy, which was93.3%, which also indicated the potential of the dimension reduction.