Application of Gamma Correction and MobileNet Methods for Leaf Image Classification

Abstract

This study proposed an enhanced leaf image classification model by integrating gamma correction as a preprocessing technique with the MobileNet (MNET) architecture to improve visual feature extraction. The dataset consisted of 750 images representing five classes of medicinal plants, namely Psidium guajava, Syzygium polyanthum, Piper betle, Annona muricata, and Andrographis paniculata, obtained from personal documentation, online sources, and public datasets. Gamma correction was applied to adjust illumination and enhance leaf texture clarity, followed by resizing and normalization processes. Data augmentation was performed using rotation, contrast adjustment, horizontal and vertical flipping, brightness adjustment, and channel shifting to increase training data variation. The MobileNet architecture was expanded with additional layers, including global average pooling, flatten, Dense–ReLU, and Dense–softmax, enabling it to function as an efficient feature extractor and classifier. Experiments were conducted using a batch size of 32, 50 epochs, the Adam optimizer, and a learning rate of 0.0001. The combined MNET and gamma correction model achieved a training accuracy of 99.00%, a validation accuracy of 87.50%, and a testing accuracy of 84.16%.