A Lightweight Convolutional Neural Network for Apple Leaf Disease Classification Using Data Augmentation

Abstract

Apple leaf diseases, particularly scab and rust, significantly reduce fruit yield and quality; therefore early diagnosis is crucial for effective crop management. Many nations grow apples for their nutritious and economic wealth. These diseases harm plant leaves restrictive photosynthesis and health. Occasionally illnesses spread quickly in agricultural areas, causing significant economic losses for farmers. Many rural and distant agricultural areas lack expert agricultural help making disease analysis challenging. Manual inspection usually misdiagnoses and delays action for diseases with identical indications. Although transfer learning (TL) models ResNet, DenseNet, and EfficientNetB0, VGG16, MobileNetV2 have achieved fine and need heaps of memory and processing. Their computational requirements make deployment difficult on low-cost devices. To address this limitation, this study develop a lightweight, effectual CNN model trained from scratch for classifying apple leaf images into four categories: healthy, rust, scab, and multiple diseases. The proposed model was developed from scratch using the TensorFlow and Keras frameworks. Its architecture consists of several convolutional layers, batch normalization layers, activation functions, max pooling layers, ReLU activation, global average pooling, dense layers and dropout regularization. During training, rotating, flipping, zooming and contrast adjustment are used to make the model better at generating output and reduce overfitting. Apple leaf images from Plant Pathology 2020 are used in this study. It achieves an overall classification accuracy of 93.61%, with class-wise F1-scores ranging from 0.70 to 0.81. These results demonstrate that the proposed CNN achieves good classification performance while maintaining a simple architecture suitable for identifying apple leaf diseases in the context of precision agriculture.