A Comparative Analysis of Machine Learning and Deep Learning Techniques for Leaf Vein Examination

An automated plant species identification system can significantly assist botanists and laypersons in rapidly and accurately identifying plant species. Deep learning has proven to be a robust tool for feature extraction, offering deeper insights into the structural and morphological details of images. In this research, a hybrid approach integrating CNN-based feature extraction with machine learning classification is explored. Leaf vein patterns are processed using Sobel-based segmentation techniques to extract vein morphometric features, which were compared against CNN-extracted features. Three pre-trained CNN models, namely VGG16, ResNet50, and InceptionV3, were used to extract hierarchical feature representations that are classified using deep learning and machine learning models. The comprehensive CNN models achieved competitive accuracies, with InceptionV3 surpassing 90 % with 30 epochs, while VGG16 followed a similar trend with accuracy improvement across epochs. However, the CNN-ML hybrid approach demonstrated superior performance, with Random Forest trained on ResNet50 features achieved the highest accuracy of 92.04 %. The segmentation methods used, focusing on simple pixel-based and edge-detection techniques, achieved lower accuracy at 67.23 %, underscoring the advantage of CNN-based feature extraction. The study highlights that CNN extracted features with robust classifiers like Random Forest, outperform traditional vein analysis techniques that rely on manually segmented vein patterns and basic geometric calculations and standalone deep learning models.