Feature extraction and classification application with deep learning

In botany and agriculture, leaf classification is a critical process that provides vital information for studies on biodiversity, ecological research, and plant species identification. Pre-existing datasets facilitate the development and evaluation of advanced classification algorithms by offering a comprehensive collection of leaf images from various plant species. This study presents a robust methodology for classifying leaf images through feature extraction and classification stages. The created dataset contains a total of 15,000 images, with 1,000 images per class. Features are extracted using pre-trained neural networks. No preprocessing methods were used. The aim is to reduce dimensionality, eliminate irrelevant or unnecessary features, and improve data quality by systematically integrating certain techniques. When dealing with large feature sets and a large number of classes, improving classification accuracy requires a combination of data preprocessing, model selection, regularization techniques, and fine-tuning. Since the dataset was low-profile, this was not necessary in this study. When the dataset is scaled up in future studies, it will be necessary to use some preprocessing methods to ensure that the attributes are more robust. The results show that when the training rate exceeded 40% in the study where two models were used for feature extraction on a 15-class dataset, the classification performance exceeded 90% for each model.