Automated Multi-Class Classification of Plant Seedling Species Using Deep Convolutional Neural Networks

Timely and accurate identification of plant seedlings is important for precision agriculture, particularly for early crop monitoring and weed management. Manual seedling identification is often time-consuming, subjective, and prone to error, especially when plant species exhibit similar visual characteristics at early growth stages. This study presents an application-oriented deep learning framework based on a compact four-layer custom Convolutional Neural Network (CNN) for multi-class classification of plant seedling images. The publicly available Kaggle Plant Seedlings Classification dataset, consisting of 12 seedling species, was used for model development and evaluation. To improve model generalization and reduce overfitting, systematic image preprocessing and data augmentation techniques were applied during training. The proposed CNN achieved an overall classification accuracy of 90%, a validation loss of 0.3837, and a macro-averaged F1-score of 0.89. The use of data augmentation improved validation accuracy from 78% to 90%, demonstrating its positive effect on classification performance. Class-wise evaluation using precision, recall, F1-score, and confusion matrix analysis showed strong performance for visually distinctive species such as Charlock, Common Chickweed, and Sugar Beet, while most misclassifications occurred among morphologically similar classes. The main contribution of this study lies in developing and empirically evaluating a simple, reproducible, and computationally efficient CNN-based pipeline for plant seedling classification. The results indicate that a compact CNN combined with appropriate preprocessing and augmentation can provide practical performance for automated plant monitoring applications, while also highlighting the need for more advanced feature extraction methods to improve classification of visually similar species.