Smart System for Mushroom Identification and Classification: A Systematic Literature Review

Abstract

Background: Mushrooms play an essential role in ecological balance and significantly contribute to economic activities. However, the close visual resemblance between edible and toxic species continues to cause fatal poisoning incidents worldwide. Conventional identification approaches rely heavily on expert judgment, making them subjective, time-consuming, and unsuitable for large-scale or real-time use. Recent developments in artificial intelligence (AI), particularly deep learning techniques, have created new opportunities for developing automated and reliable identification systems. Objective: This study systematically reviews recent research on intelligent technologies for mushroom identification and classification. This study maps the main research objectives, data types, morphological features, technologies, algorithms, evaluation practices, and remaining challenges, with particular attention to toxicity classification and safety-critical decision support. Methods: A systematic literature review was conducted following the PRISMA 2020 protocol and the PICOC framework. We screened publications indexed in Scopus between 2021 and 2025. From 1,308 initial records, 90 high-quality studies were selected for detailed analysis. Results: The analysis of the selected studies reveals that the majority of research focuses on discriminating between edible and poisonous mushrooms (52 studies), predominantly using visual characteristics related to shape (64 studies) and color (59 studies). Deep learning techniques, especially convolutional neural networks (CNNs) (40 studies) and vision transformers (11 studies), dominate the field and frequently report classification accuracies exceeding 95%. Despite these achievements, several challenges persist, including the difficulty of fine-grained classification among visually similar species, limited dataset availability, and performance degradation in complex natural environments. Conclusion: Although AI-based approaches have considerable potential to support mushroom identification, their reliability is limited by reliance on visual data alone when species are morphologically similar. Future studies should place greater emphasis on lightweight models for field deployment, multimodal sensing, risk-aware evaluation, and Explainable AI (XAI) so that intelligent systems can be used more safely and transparently in real-world contexts. Keywords: Mushroom Identification, Artificial Intelligence, Deep Learning, Food Safety, Systematic Literature Review