Education at Lviv Polytechnic

Intelligent multimodal web system for automated identification and classification of mushrooms

Students Name: Anastasyn Ihor Andriiovych

Qualification Level: magister

Speciality: Information Technology Design

Institute: Institute of Computer Science and Information Technologies

Mode of Study: full

Academic Year: 2025-2026 н.р.

Language of Defence: ukrainian

Abstract: Anastasyn I.A., (Supervisor) Artyshchuk I.V. “Intelligent multimodal web system for automated identification and classification of mushrooms” Master’s Qualification Thesis. – National University “Lviv Polytechnic”, Lviv, 2025. Extended Abstract Relevance of the study. Modern advances in computer vision and artificial intelligence open up new opportunities for automated identification of natural objects, particularly mushrooms. Identifying mushrooms from images remains a challenging task even for experienced mycologists, as many species share similar morphological features. At the same time, the increasing number of poisoning cases highlights the urgent need for reliable digital tools capable of distinguishing edible and poisonous species. The development of the mobile application Fungi.Info, equipped with an embedded neural network for mushroom classification, represents a relevant and promising direction that integrates artificial intelligence, multimodal data processing, and digital biology education. The aim of the thesis is to design and implement an intelligent system for automatic mushroom identification from images, combining deep learning techniques, optimized neural network models, and mobile technologies to ensure accurate, fast, and safe classification in field conditions. The object of study is the process of image recognition and classification of mushrooms using deep learning models. The subject of study includes computer vision methods, neural network architectures, and the principles of integrating TensorFlow Lite models into the Android mobile environment. Structure and content of the work The first chapter presents an analysis of mushrooms as biological organisms, describing their morphological characteristics, taxonomic structure, and species diversity. Existing computer vision approaches and automated identification systems—such as iNaturalist, Mushroom Observer, and PlantNet—are reviewed. A comparative analysis revealed that most current systems lack offline functionality and have limited classification accuracy. The second chapter provides a comprehensive system analysis, formulating functional, accuracy, and usability requirements for the future application. The structure of a relational database is designed to store information about species, toxins, users, and classification results. A three-tier architecture is proposed, comprising the data module, artificial intelligence module, and user interface module. The third chapter describes the implementation of the mobile application Fungi.Info, which performs image capture, preprocessing, and inference directly on the user’s device. The deep learning model, based on TensorFlow MobileNetV3-Large, was trained on open datasets (Kaggle, iNaturalist, FGVCx Fungi), optimized using FP16 quantization, and converted into the TensorFlow Lite format for on-device inference. The user interface was designed in Figma following Material Design Components, ensuring a clean and intuitive layout. Testing results demonstrated 91% classification accuracy and stable offline operation. The thesis consists of an introduction, three chapters, conclusions, a list of references, and appendices. It includes 25 figures, 10 tables, 5 code fragments, and several visual illustrations of the system architecture, interface, and training performance graphs. KEYWORDS: artificial intelligence, deep learning, computer vision, TensorFlow Lite, MobileNetV3, Android application, mushroom classification, offline identification, environmental monitoring, Fungi.Info. References: 1. Chollet, F. Deep Learning with Python. Manning Publications, 2021. – 504 p. 2. TensorFlow Lite Documentation. Google Developers. [Online resource]. URL: https://www.tensorflow.org/lite (accessed: 08.04.2025). 3. iNaturalist Platform. [Online resource]. URL: https://www.inaturalist.org (accessed: 08.04.2025). 4. FGVCx Fungi Image Dataset. [Online resource]. URL: https://www.kaggle.com/competitions/fgvcx-fungi (accessed: 08.04.2025).