Deep feature representation for automated plant species classification from leaf images

Abstract

Automated plant species classification using leaf images holds immense potential for advancing agricultural research, biodiversity conservation, and ecological monitoring. This study introduces a novel approach leveraging deep feature representation to achieve accurate and efficient classification based on leaf morphology. Convolutional neural networks (CNNs), including VGG16, ResNet50, DenseNet1, Inception, and Xception, are employed to extract high-level features from leaf images, capturing intricate patterns essential for species differentiation. To manage the extensive feature set extracted by these models, optimization techniques such as principal component analysis (PCA), variance thresholding, and recursive feature elimination (RFE) are applied. These methods streamline the feature set, making the classification process more efficient. The optimized features are then trained using classifiers like support vector machine (SVM), k-nearest neighbors (K-NN), decision trees (DT), and naive Bayes (NB), achieving average accuracies of 98.6%, 96.6%, 99.6%, and 99.7%, respectively, across various cross-validation methods. Experimental results on benchmark datasets demonstrate the effectiveness of this approach, achieving state-of-the-art performance in plant species classification. This work underscores the potential of deep feature representation in automated plant species classification, offering valuable insights for applications in agriculture, ecology, and environmental science.