Symptom based automated detection of citrus diseases using color histogram and textural descriptors

This paper presents a technique to detect and classify major citrus diseases of economic importance. Kinnow mandarin being 80% of Pakistan citrus industry was the main focus of study. Due to a little variation in symptoms of different plant diseases, the diagnosis requires the expert's opinion in diseases detection. The inappropriate diagnosis may lead to tremendous amount of economical loss for farmers in terms of inputs like pesticides. For many decades, computers have been used to provide automatic solutions instead of a manual diagnosis of plant diseases which is costly and error prone. The proposed method applied ΔE color difference algorithm to separate the disease affected area, further, color histogram and textural features were used to classify diseases. Our method out performed and achieved overall 99.9% accuracy and similar sensitivity with 0.99 area under the curve. Moreover, the combination of color and texture features was used for experiments and achieves similar results, as compared to individual channels. Principle components analysis was applied for the features set dimension reduction and these reduced features were also tested using state of the art classifiers.