Detection Methods Of Fruit Maturity And Diseases Based On Image And Spectral Techniques

Fruit detection during different mature stages,and fruit diseases detection help growers to manage each crop production on a site-specific basis to reduce waste,raise profits,and maintain the quality of the environment.With today's increasing competition,early and accurate yield forecasting of immature green fruit and blueberry fruit in different maturity stages ahead of harvesting time helps growers to identify site-specific growth conditions of trees at an earlier stage so that they can properly plan application of nutrients or fertilizers during the fruit immaturity stages.Yield mapping can also help competitive farmers to promote crop yield while minimizing costs by determining how much labour would be needed during the harvesting period and well allocate labour depending on the yield prediction in advance.Recently,Citrus Black Spot,CBS,has become one of the devastating diseases for citrus growers.To control the spread of the disease and reduce the risk of infection,it is better to detect the disease before harvest time,at least,before being transported to other areas with no disease yet.Many researchers studied above-mentioned issues and already made plenty of breakthroughs,there are still many remained problems need to be solved.Based on the previous research results and remained problems,this dissertation conducted study on immature green citrus detection,citrus black spot disease detection and identification of different symptoms,and blueberry fruit detection under different maturity.In general,the major work,results and contributions of this dissertation are as follows.(1)Recognition and detection of green immature citrus fruit more accurately and efficiently in groves under natural illumination conditions provides a promising benefit for growers to plan application of nutrients during the fruit maturing stages and estimate their yield and profit prior to harvesting period.The goal of this study was to develop a robust and fast algorithm to detect and count immature green citrus fruit in individual trees from colour images acquired with different fruit sizes and under various illumination conditions.Adaptive Red and Blue chromatic map(ARB)was created and combined with the Hue image extracted after histogram equalization(HEH).Sum of absolute transformed difference(SATD),a block-matching method,was applied to detect potential fruit pixels.After OR operation of the results obtained from colour and SATD analysis which kept as many fruit pixels as possible,a kernel support vector machine(SVM)classifier was built to remove false positives based on five selected texture features.The algorithm was evaluated with a set of testing images,and achieved more than 83%recognition accuracy.The proposed method can provide a more efficient way for green citrus identification in a grove using colour images.(2)Citrus black spot(CBS),one of the most common fungal diseases of citrus,causes lesions on the rind and early fruit drop before its mature stage.This disease can significantly reduce crop yield,making blemished fruit unsuitable for market.A portable USB2000+spectrometer was used to acquire spectra reflectance of citrus fruit in the lab with the wavelength range from 340 nm to 1030 nm.To reduce the data dimensionality and select the useful bands for further application,principal components analysis(PCA)and four feature ranking methods,T-test,Kullback-Leibler divergence,Chernoff bound and Receiver Operating Characteristic(ROC)were applied.One important wavelength,525 nm,was selected and used to classify healthy and CBS infected.Sequential minimal optimization(SMO),Radical basis function network(RBF),and C4.5 classification methods were used to evaluate the performance of the selected wavelengths,and SMO achieved the highest accuracy of 99.37%.In order to compare the performance of classification accuracies according to optimal wavelengths selected using different methods,two other methods,Sequential floating forward selection(SFFS)and Mutual information(MI),were applied.Two wavelengths,527 nm and 917 nm,were selected based on SFFS,while the MI method selected 513-531 nm as the optimal wavelength range,and the highest recognition accuracy was 99.06%,which was lower than that of using 525 nm.Then SFFS was applied to find the optimal wavelengths for further distinguishing three CBS symptoms.C4.5 method was used to evaluate the performance of distinguishing CBS infected and healthy based on selected wavelengths and the highest overall classification accuracy was 73.77%.(3)In order to detect CBS in the grove before harvesting time,a hyperspectral imaging system was established and the spectral signature of healthy and infected citrus fruit were studied with the wavelength range from 396 nm to 1010 nm,to identify diseased fruit from healthy ones.However,hyperspectral images contain hundreds of wavelengths,and many of them would be considered as redundant,which may even decrease the classification accuracy.In this study,to reduce the dimensionality of hyperspectral images and select the useful bands for further application principal components analysis(PCA)and four band ranking methods,T-test,Kullback-Leibler distance,Chernoff bound and Receiver Operating Characteristic(ROC)were applied.Two important wavelengths,481 nm and 534 nm,were selected to classify healthy and CBS infected samples with three main symptoms,Hard spot,Cracked spot and Virulent.Kernel Support vector machine(KSVM),K-nearest neighbor(KNN)and Radical basis function network(RBF)classification methods were used to evaluate the performance of the selected bands,and KSVM achieved the highest accuracy of 98.41%.Then two feature selection methods,Sequential Forward Selection(SFS)and Random Subset Feature Selection(RSFS),were applied to find the optimal wavelengths and to classify cracked spot from other symptoms.Random forest and RBF network were used to test the performance of distinguishing CBS infected and healthy based on selected bands.The highest overall classification accuracy was 84.62%using RBF,which was a little higher than using Random Forest.The recognition accuracy for hard spot was about 90.63%,and the other two symptoms,virulent and cracked spot achieved the highest classification rate of 84.07%and 74.05%,respectively.(4)The blueberry industry has been increasingly important to both Florida and United States economically and yielding high profits in the fresh market.However,the mature period is relatively short,and the price after that period drops dramatically.Harvesting of blueberry is very labor-intensive because most of the fruits are handpicked.Therefore,early estimation of blueberry fruit yield in a large scale blueberry farm is crucial for the market and for labor planning in order to reduce harvest expense and increase profits.The objectives of this study were to investigate the feasibility of blueberry detection based on color images and multispectral image,and to build robust classification models,which tolerates outdoor illumination changes and complicated background information.In this study,there are three growth stages,mature,near mature,and young.Color features of different growth stages were studied and used with image fusion method using RGB images.B-G fusion combined with Hue component extracted mature and near mature fruit from young and background.Then B-R was applied to further classify mature from near mature fruit.In order to segment young fruit from the background,three optimal color features,B,H,and V components,were selected from twelve features.KNN,NB,C4.5 and Random Forest were established with 10 fold cross validation to identify young fruit.Results showed the identify accuracies of mature and near mature fruit using C4.5 were 89.22%and 85.61%,respectively.The young fruit achieved the lowest accuracy for about 69%.For the multispectral imaging,vegetation index,NDVI,was used to remove non-vegetation regions.Meanwhile,according to the reflectance performance of different growth stages,young and near mature were extracted from mature and leaves.Color features were studied and used to further classify three growth stages.Results showed that the highest recognition rate was 86.52%of young fruit using C4.5.It can be seen that mature and near mature fruit achieved higher accuries using RGB image while young fruit has better identification rate using multispectral imging.Results of this study could be used to recognize blueberry fruit with different growth stages and further used for a blueberry yield mapping system for large-scale blueberry farms based on color images.