Abstract:
"The recognition and classification of herbal plants in Sri Lanka represent a significant challenge due to the variability in natural environments and the diverse array of plant species. This project aimed to develop an application utilizing deep learning techniques to address this challenge. The application's primary goal is to provide users with accurate identification of herbal plants and valuable information regarding their medicinal properties. Leveraging convolutional neural network (CNN) models and transfer learning, the project aimed to overcome complexities associated with plant classification in natural environments, including variations in lighting conditions and background diversity.
The methodology involved several critical steps in solving the problem of plant classification. Initially, a comprehensive dataset comprising images of herbal plants commonly found in Sri Lanka was collected and curated to ensure representation across diverse species and environmental conditions. Subsequently, state-of-the-art deep learning architectures tailored for image classification tasks were employed to train robust CNN models capable of accurately identifying plant species from input images. Additionally, rigorous data preprocessing techniques, including image scaling and augmentation, were implemented to enhance the model's generalization capabilities and improve its performance under varying conditions.
In evaluating the initial results of the project, extensive testing and validation procedures were conducted to assess the model's classification accuracy and performance. Utilizing confusion matrices and precision-recall curves, quantitative analysis revealed an accuracy of 94%, indicating promising levels of accuracy and reliability. These results suggest that the developed application holds significant potential as a valuable tool for herbal plant recognition and medicinal information dissemination. Moving forward, further refinement and optimization efforts will focus on enhancing the model's robustness and generalization capabilities to ensure its effectiveness across diverse environmental conditions and plant species."
