PSIII-2 Machine learning-based image classification of sericea lespedeza: A comparative study.

Abstract The drought-resistant leguminous plant Sericea lespedeza (Lespedeza cuneata) is highly valued for its soil conservation, livestock feeding, and sustainable agriculture benefits. Species classification is essential for accurate identification and suitable use in various agroecosystems. This research makes a unique contribution by using machine learning-based image classification methods to categorize six distinct species of Sericea lespedeza. Three different classification models, such as Support Vector Machine (SVM), Backpropagation Neural Network (BPNN), and Convolutional Neural Network (CNN), were employed to evaluate their efficacy in species identification based on picture characteristics. High-resolution images of all six Sericea lespedeza species were captured using a Samsung A54 smartphone in a controlled environment to ensure that the lighting, angle, and focus were consistently maintained. The photos were preprocessed using data argumentation techniques to improve the model’s generalization ability, including scaling, normalization, and augmentation. For enhanced feature extraction and classification, we utilized Backpropagation Neural Networks (BPNN) and Convolutional Neural Networks (CNN), with Support Vector Machine (SVM) acting as the baseline classifier. With a weighted F1-score of 0.82, the SVM model achieved an accuracy of 82%. A weighted F1-score of 0.82 and an accuracy of 82% were the results shown by the BPNN model, which was equivalent in performance. Despite the differences, all models showed impressive classification accuracy, with high recall and precision for most species. The balanced dataset and clearly separated features among species likely contributed to the perfect classification agreement, as all of the models’ Cohen’s Kappa values were 1.0. According to these findings, machine-learning approaches can automate the classification of Sericea lespedeza species. Although SVM and BPNN showed better accuracy, CNN’s ability to learn hierarchical visual characteristics suggests that further enhancements could enhance performance with bigger datasets and more complicated structures. Improving species separation should be the focus of future research that combines hyperspectral imaging with deep learning techniques, such as transfer learning. The potential of this research to develop an app that could identify Sericea lespedeza species in real time is significant for researchers, farmers, and environmentalists, offering a practical tool for their work. This research underscores the growing importance of AI in agriculture and its potential to revolutionize farming practices.