Comparing Deep Learning (CNN) Features with Hybrid Models (HOG_CNN, LBP_CNN) for Classification of Leafs

Ayurvedic physicians use various leaves and plants medicinally, leveraging their consistent availability throughout the year to treat conditions such as skin disorders, colds, and memory enhancement. Leaves play a crucial role in plant identification, especially as fruits, seeds, and flowers often vary with the seasons. Leaf detection also aids forest management, remote sensing, and ecological assessments, providing data for species monitoring, vegetation mapping, and climate modeling. This study evaluates three methods for leaf classification: CNN (Convolutional Neural Network), LBP _CNN (integrating Local Binary Pattern with CNN), and HOG_CNN (combining Histograms of Oriented Gradients with CNN). Results indicate that while CNN alone is time-consuming and less accurate, integrating it with traditional feature extraction methods (LBP and HOG) significantly improves performance. Among the models, HOG_CNN achieved the highest accuracy with reduced processing time compared to standalone CNN. The dominance of the proposed models can be ascertained by measuring accuracy with various image sizes in contrast to other cutting-edze techniques.