روش ﻣﺒﺘﻨﯽ ﺑﺮ ﭘﺮدازش ﺗﺼﻮﯾﺮ ﺑﻪ ﻣﻨﻈﻮر ﺗﺸﺨﯿﺺ ﺧﻮدﮐﺎر ﺑﯿﻤﺎري ﺑﺮگ درﺧﺖ اﻧﮕﻮر

Image Processing Based Method for Automatic Detection of Grape leaf Diseases

ﺗﺸﺨﯿﺺ ﺳﺮﯾﻊ و ﭘﯿﺸﮕﯿﺮي از ﮔﺴﺘﺮش ﺑﯿﻤﺎري ﻣﺤﺼﻮﻻت ﮐﺸﺎورزي، ﻣﯽﺗﻮاﻧﺪ ﺗﻠﻔﺎت ﻣﻘﺎﺑﻠﻪ ﺑﺎ ﺑﯿﻤﺎري را ﺑﻪ ﻣﯿﺰان ﻗﺎﺑﻞ ﺗﻮﺟﻬﯽ ﮐﺎﻫﺶ دﻫﺪ. در اﯾﻦ ﭘﮋوﻫﺶ، ﺳﺎﻣﺎﻧﻪاي ﻫﻮﺷﻤﻨﺪ ﺑﺮ ﻣﺒﻨﺎي ﭘﺮدازش ﺗﺼﻮﯾﺮ ﺑﺮاي ﺗﺸﺨﯿﺺ ﺑﯿﻤﺎريﻫﺎي ﺑﺮگ درﺧﺖ اﻧﮕﻮر )Sultana - Vitis vinifera اراﺋﻪ ﮔﺮدﯾﺪه اﺳﺖ. ﺑﺪﯾﻦ ﻣﻨﻈﻮر، وﯾﮋﮔﯽﻫﺎي ﻣﺨﺘﻠﻒ ﺑﺎﻓﺖ ﺗﺼﻮﯾﺮ از ﻫﯿﺴﺘﻮﮔﺮام ﺳﻄﺢ ﺧﺎﮐﺴﺘﺮي )GLH(، ﻣﺎﺗﺮﯾﺲ ﻫﻢ-رﺧﺪاد ﺳﻄﺢ ﺧﺎﮐﺴﺘﺮي )GLCM(، ﻣﺎﺗﺮﯾﺲ ﻃﻮل ﺑﺮدار ﺳﻄﺢ ﺧﺎﮐﺴﺘﺮي )GLRM( و اﻟﮕﻮي دودوﯾﯽ ﻣﺤﻠﯽ )LBP( اﺳﺘﺨﺮاج ﺷﺪ. ﺑﺮاي ﻣﺪلﺳﺎزي وﯾﮋﮔﯽﻫﺎ، از دو ﻣﺪل ﺷﺒﮑﻪ ﻋﺼﺒﯽ )ANN( و ﻣﺎﺷﯿﻦ ﺑﺮدار ﭘﺸﺘﯿﺒﺎن )SVM( اﺳﺘﻔﺎده ﺷﺪ. ﭘﺎﯾﮕﺎه داده ﻣﻮرد اﺳﺘﻔﺎده، ﻣﺘﺸﮑﻞ از 4062 ﺗﺼﻮﯾﺮ، ﺷﺎﻣﻞ ﺑﺮگ ﺳﺎﻟﻢ، ﻣﺒﺘﻼ ﺑﻪ ﭘﻮﺳﯿﺪﮔﯽ ﺳﯿﺎه، اﺳﮑﺎ و ﻟﮑﻪ اﯾﺰارﯾﻮﭘﺴﯿﺲ اﺳﺖ. ﻧﺘﺎﯾﺞ ﻧﺸﺎن دادﻧﺪ ﮐﻪ ﻣﺪل SVM ﺑﺎ اﺳﺘﻔﺎده از وﯾﮋﮔﯽﻫﺎي GLRM ﺑﺎ ﻣﺘﻮﺳﻂ دﻗﺖ 89/70% ﺑﻬﺘﺮﯾﻦ ﻋﻤﻠﮑﺮد را از ﺧﻮد ﻧﺸﺎن داد. ﻫﻤﭽﻨﯿﻦ ﻧﺘﺎﯾﺞ ﻧﺸﺎن دادﻧﺪ، اﺳﺘﻔﺎده از ﺗﻤﺎم وﯾﮋﮔﯽﻫﺎي اﺳﺘﺨﺮاج ﯾﺎﻓﺘﻪ ﺑﻪ ﺻﻮرت ﺑﺮدار وﯾﮋﮔﯽ واﺣﺪ، اﻓﺰاﯾﺶ دﻗﺖ دﺳﺘﻪﺑﻨﺪي را ﺑﻪ دﻧﺒﺎل دارد. ﻣﺪل ANN SVM ﺑﺎ اﺳﺘﻔﺎده از ﺗﻤﺎم وﯾﮋﮔﯽﻫﺎ ﺑﺘﺮﺗﯿﺐ ﺑﺮاي دادهﻫﺎي آﻣﻮزﺷﯽ دﻗﺖ 95/04 ،%91/10 % و ﺑﺮاي دادهﻫﺎي آزﻣﻮن ﻣﯿﺰان دﻗﺖ 89/93% و 91/75% را ﻧﺘﯿﺠﻪ دادﻧﺪ. در ﻧﻬﺎﯾﺖ، ﺑﺎ اﺳﺘﻔﺎده از اﻟﮕﻮرﯾﺘﻢ ﮐﻠﻮﻧﯽ زﻧﺒﻮر ژﻧﺘﯿﮑﯽ )GBC( و ﮐﺎﻫﺶ ﺗﻌﺪاد وﯾﮋﮔﯽﻫﺎ ﺑﻪ 34 و 46 ﺑﻪ ﺗﺮﺗﯿﺐ ﺑﺮاي ﻣﺪلﻫﺎي ANN و SVM ﻣﯿﺎﻧﮕﯿﻦ دﻗﺖ 97/20% و 94/10% ﺑﺮاي آﻣﻮزش و آزﻣﻮن ﻣﺪل ANN و 93/01% و 92/33% ﺑﺮاي آﻣﻮزش و آزﻣﻮن ﻣﺪل SVM ﺑﻪ دﺳﺖ آﻣﺪ ﮐﻪ ﻧﺸﺎن دﻫﻨﺪه ﺑﻬﺒﻮد ﻧﺘﺎﯾﺞ ﺗﻮﺳﻂ اﻟﮕﻮرﯾﺘﻢ GBC ﻣﯽﺑﺎﺷﺪ. روش ﭘﯿﺸﻨﻬﺎدي در ﺗﺸﺨﯿﺺ ﺑﯿﻤﺎريﻫﺎي ﺑﺮگ اﻧﮕﻮر ﮐﺎرآﻣﺪ ارزﯾﺎﺑﯽ ﺷﺪ.

Rapid detection and prevention of disease spread in agricultural products can significantly reduce losses and costs of disease control. In this study, an intelligent system based on image processing method has been presented for detection of grape (Sultana - Vitis vinifera) leaf diseases. For this purpose, different image texture features were extracted from the Gray Level histogram (GLH), Gray Level Co-occurrence Matrix (GLCM), Gray Level Run Length Matrix (GLRM) and Local Binary Pattern (LBP) algorithms. Two models of Artificial Neural Network (ANN) and Support Vector Machine (SVM) were used to model the features. The dataset consists of 4062 images including healthy leaves, Black Rot, Esca and Isariopsis leaves. The results showed that the SVM model based on GLRM features with an average accuracy of 89.70% showed the best performance. The results also showed that the use of all extracted features as a single feature vector increases the accuracy of classification. The accuracy of the SVM and ANN models using all of the features for training data were 91.10%, 95.04%, and for the test data were 89.93% and 91.75%, respectively. Finally, using Genetic Bee Colony (GBC) algorithm and reducing the number of features to 34 and 46 for ANN and SVM models, respectively, the average accuracy of 97.20% and 94.10% for training and testing of ANN model and 93.01% and 92.33% for training and testing of SVM model were obtained, which shows the improvement of results by GBC algorithm. The proposed method was evaluated as efficient in diagnosing grape leaf diseases.