Image Enhancement using Haar Wavelet for Image Classification of Diseases on Strawberry Leaves with Convolutional Neural Networks

Abstract

The classification of diseases on strawberry leaves is interesting and important in agriculture. It can help enhance the efficiency of disease management and pesticide management for diseases such as leaf spot, leaf blight, and powdery mildew. The photographs need to be clear because each disease on the strawberry leaves has distinct characteristics. This article proposes a method to improve the efficiency of strawberry leaf disease classification with Convolutional Neural Networks (CNNs), using the Haar wavelet for image enhancement. A total of 2,192 images of strawberry leaves were used, and the results were compared with images that did not use the Haar wavelet method. Training is conducted for 300 and 500 epochs. The experimental results showed that the proposed method increased accuracy from 94.61% to 96.08% and the F1-Score from 94.89% to 96.33% for 300 epochs, using a processing time of 36 minutes, and from 95.59% to 97.44% and the F1-Score from 95.65% to 97.63% for 500 epochs, using a processing time of 62 minutes.