Login

Proceedings

Find matching any: Reset
Add filter to result:
Strawberry powdery mildew detection using color co-occurrence matrix based machine vision algorithm
M. Mahmud, Y. Chang, B. Prithiviraj
Dalhousie University

Computer vision systems have been utilized to develop decision support system for taking strategic decision on the agriculture protection research. However, strawberry powdery mildew disease scrutiny is still manually carried out due to lack of technological development for plant disease detection task. Image processing is considered as one of the major area for disease detection in agricultural crop cultivation. Therefore, present study proposed an image processing technique used to detect and classify fungal powdery mildew disease on strawberry cultivars. Two experimental sites were utilized in western Nova Scotia for image data collection. All collected images were segmented based on a certain threshold value which were utilized for textural features extraction thereafter. The method used in the study consists of two main stages: a color co-occurrence (CCM) based textural feature extraction and a decision-making process using classifiers. Textural feature extraction started with the conversion of original red green blue (RGB) image into luminance, hue, saturation and intensity images. Total forty-four features were extracted using CCM textural analysis. A graphical user interface for textural analysis was developed using C# programming language. In the decision-making process, the algorithm makes use of the decision tree and multiple regression classifier to classify powdery mildew disease leaves. The performance of developed powdery mildew detection algorithm was tested and examined based on classification accuracy. Internal and external validation were experimented from the normalized data of two fields, having 300 images from healthy and diseases leaves. Results of the classifier development showed that decision tree classification on different textural features produced agreeable results, where healthy and powdery mildew leaves were classified. The algorithm was able to achieve overall 86% accuracy for powdery mildew disease leaf recognition. Therefore, this study can help to detect powdery mildew and allow strawberry growers to take appropriate and effective action to control powdery mildew disease.

Keyword: Powdery mildew, Image segmentation, Textural analysis, Color co-occurrence matrix, Decision tree, multiple regression