Development of AI-based Image Recognition for Automated Plant Species Identification

Thesis Abstract

Accurate identification of plant species is fundamental to biodiversity conservation, ecological research, and sustainable forestry management, yet traditional manual classification methods are time-consuming, subject to human error, and often inaccessible to non-experts. This study aims to develop an advanced artificial intelligence (AI)-based image recognition system to automate plant species identification, enhancing speed, accuracy, and scalability of botanical data collection. The specific objectives are to (1) compile a comprehensive repository of plant images across diverse taxonomic groups; (2) design and train a convolutional neural network (CNN) model optimized for plant feature recognition; (3) evaluate the model's classification accuracy against expert identifications; (4) implement the system within a mobile application framework for field use; and (5) assess user performance and system reliability in real-world conditions. The research adopts a quantitative methodological approach, integrating supervised machine learning with experimental validation. The population comprises a diverse selection of plant species representatives from regional botanical collections, with a sample size of approximately 10,000 high-resolution images covering at least 1,000 species across different families. Stratified random sampling ensures representation across ecological zones and plant morphology types. Data collection involves capturing images under standardized conditions from botanical gardens and field sites, complemented by expert-verified labels to serve as ground truth. The primary instrument for data collection is a digital camera system, alongside a custom mobile application designed for capturing and pre-processing images. For model training and validation, the dataset is partitioned into training (70%), validation (15%), and testing (15%) subsets to prevent overfitting. The image recognition model is developed using TensorFlow and Keras frameworks, employing a transfer learning approach with pre-trained models such as InceptionV3 and ResNet50. Hyperparameter tuning is conducted through grid search techniques to optimize model performance. The model’s effectiveness is evaluated using metrics such as accuracy, precision, recall, and F1-score, with confusion matrices providing insight into classification errors. Statistical analysis of the model's performance against baseline manual identification, validated via Cohen’s kappa coefficient, assesses the reliability of the automated system. Additionally, usability testing of the mobile application incorporates quantitative user performance metrics and qualitative feedback. Expected findings suggest that the AI-driven recognition system can achieve classification accuracies exceeding 90%, significantly outperforming traditional visual identification, especially for species with subtle morphological differences. The system is anticipated to demonstrate robustness across different environmental conditions and image qualities, establishing its viability for field application. The integration within a mobile platform is expected to enhance accessibility for researchers, conservationists, and citizen scientists, thereby broadening the scope of botanical data collection and monitoring. This research contributes novel insights into the application of deep learning techniques for plant taxonomy, demonstrating an effective AI framework that combines high accuracy with operational practicality in diverse settings. It addresses gaps in existing literature by providing a comprehensive evaluation of neural network models specialized for botanical imagery and highlights the potential for scalable, automated plant identification tools in ecological and conservation disciplines. The findings support a paradigm shift toward digitized, real-time botanical surveys, reducing reliance on expert knowledge and increasing efficiency. Ultimately, the study concludes that AI-based image recognition constitutes a transformative technology for botanical sciences, with recommendations for further enhancement through integration with geographic information systems (GIS), expansion to include phenological and ecological data, and development of multilingual user interfaces to facilitate global application. Future research should explore the inclusion of multispectral imaging to improve species differentiation and incorporate community-based data collection to enrich the herbarium-style datasets further.