Multiclass Classification of Red Onion Leaf Diseases Using a Simple CNN with UBLAF Image Enhancement

Abstract

Shallots are one of the agricultural products that support the economy of an agrarian country like Indonesia. The challenge in shallot cultivation is the plant diseases that can cause a decrease in the quality and quantity of the harvest. The diseases of red onion plants can be observed from the condition of the leaves of the red onion plants themselves. This research proposes an artificial intelligence-based approach to classify red onion plant diseases by examining the condition of the red onion plant leaves. This study proposes a CNN with a simple architecture capable of classifying red onion plant diseases. Additionally, this research proposes an image processing technique to enhance the quality of red onion plant images by sharpening the image in the red onion leaf area using unsharp masking and blurring other areas using Gaussian blur with the help of Laplacian variance. The image processing technique in this study is called Unsharp-Blur Laplacian-Adaptive Filter (UBLAF). The result of this research is a CNN with a simple architecture that performs well in classifying red onion plant diseases with an accuracy of 0.96 and an F1-score of 0.96. Another result obtained is the role of UBLAF, which is able to improve the quality of features and the generalization capability of the proposed CNN model. This is evident from the performance results of the CNN model using images obtained from the UBLAF technique, which achieved an accuracy of 0.98 and an f1-score of 0.98.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint