MedLeafNet: A Feature Fusion Network Integrating Transformer and CNN Modules for Indian Medical Leaf Classification

A new feature fusion network called “MedLeafNet” is presented in this study specifically for the categorization of Indian medicinal leaves. Using a hybrid architecture, MedLeafNet integrates the best features of Convolutional Neural Network (CNN) modules for spatial feature extraction and Transformer modules for sequence modeling. Because of the complexity of classifying Indian medicinal leaves—which are famous for their varied and powerful medicinal properties—the suggested methodology is well suited to the task. For this research, 80 different types of Indian leaves were analyzed, each with its well-known therapeutic properties. To improve its capacity to detect complex patterns in leaves, Me dLeafNet incorporates Transformer and CNN modules to efficiently gather both global and local information. The suggested model was optimized using several different optimizers, such as Adam, Adadelta, Gradient Descent, AdaGrad, RMSProp, Adamax, Momentum, and Adaptive Moment Estimation. When trained with the Adam optimizer, MedLeafNet attained its maximum accuracy of 98.97%, as shown by extensive testing. MedLeafNet is also compared against some of the most well-known deep learning architectures, such as ResNet, Inception, VGG, Xception, and MobileNet. In terms of classification accuracy, the findings show that it is superior to these architectures, proving that it is effective for the difficult job of classifying Indian medicinal leaves.