Crop identification and disease classification using traditional machine learning and deep learning approaches

Abstract EN

Crop and disease classification is one of the important problems in automation of agricultural processes with multicropping method, where the field is cultivated with more than one crop.

In order to solve this classification problem, a study has been carried out in the field cultivating eggplant (Solanum melongena) and tomato (Solanum lycopersicum) using the images obtained from a mobile phone camera.

Textural descriptors, namely, contrast, correlation, energy, and homogeneity, were extracted from the gray-scale converted RGB image for crop identification, that is, tomato or eggplant, and the same descriptors were extracted from the gray-scale converted image from Hue Saturation Value (HSV) for disease classification (due to Cercospora leaf spot disease or two-spotted spider infestation).

Discriminant analysis, Naive Bayes algorithm, support vector machine, and neural network were the classification algorithms used with a resulting best accuracy of 97.61%, 95.62%, 98.01%, and 98.94% for crop identification and 86.09%, 76.52%, 86.96%, and 86.04% for disease classification, respectively.

Similarly, the application of algorithm with 6 histogram-based descriptors for health status detection resulted in an accuracy of 66.67%, 37.04%, 50%, and 72.9%, respectively.

A deep learning algorithm, namely, AlexNet, was also evaluated, which resulted in an accuracy of 100% for crop identification, 89.36% for health status detection, and 81.51% for disease classification.

Among the algorithms, AlexNet resulted in the best average accuracy of 90.29% for the above classification tasks.