Unveiling Nature's Code: Novel Approaches to Plant Leaf Feature Extraction

Plants are indispensable in supporting biodiversity by serving as both habitats and food supplies for an extensive variety of creatures, strengthening the overall welfare of ecosystems. In agriculture, they serve as a vital sustenance reservoir for humans as well as animal populations, highlighting their critical contribution to global food security. They are highly significant in healthcare since several medicinal substances are generated from them, providing remedies for a variety of diseases. Their organic compounds serve as the foundation for several medications, assisting in the fabrication of therapeutic treatments. Their leaves require feature extraction to distinguish and quantify various features that are crucial for plant identification as well as medical evaluation. These traits include leaf form, size, texture, and vein structures, which aid in species identification. Feature extraction provides a more concise and relevant set of features by extracting pertinent information from raw data. The paper proposes the application of HOG (Histogram of Oriented Gradients), Log Gabor, and GLCM (Gray Level Co-Occurrence Matrix) to extract features from 30 diverse plant species, subsequently proceeding with classification. The images undergo pre-processing which involves standardization using Kuwahara filter and subsequent normalization is achieved through the application of either PCA (Principal Component Analysis) or Sobel edge detection techniques. Following that, the PCA outputs are provided to HOG for feature extraction, the Sobel results are given to Log Gabor, and the unnormalized images are given to GLCM. The classification of HOG outcomes employs GBM (Gradient Boosting Machine) classifier, whereas SVM (Support Vector Machine) is utilized for the classifications of Log Gabor and GLCM outcomes. The primary aim is to ascertain the optimal feature extraction strategy.