COMPARATIVE ANALYSIS OF AI AND DEEP LEARNING TECHNIQUES FOR DETECTING CHEMICALLY RIPENED FRUITS, CHALLENGES AND FUTURE DIRECTIONS

Abstract

Automatic detection of chemically ripened fruits has become an active area of research in recent years due to rising food safety concerns and awareness regarding fruit quality degradation and health hazards caused by the misuse of artificial ripening agents. Manual techniques for detecting fruit ripeness rely on visual cues or destructive sensing, which are inherently subjective and can hardly be scaled for agricultural purposes. Several attempts have been made to leverage Artificial Intelligence (AI) & Deep Learning for automatic and non-destructive detection of fruit ripeness. In this paper, we survey current state-of-the-art methodologies while comparing their strengths & weaknesses, majorly focusing on thermal image-based approaches for the detection of chemically ripened fruits. Existing machine learning approaches like SVMs & Random Forests are computationally less expensive and are preferred in low-resource settings, while Deep Learning models, especially CNNs, have proven to perform better in terms of classification accuracy since they have the ability to automatically extract features from input data. Hybrid approaches attempting to strike a balance between accuracy and resource constraints have also shown promise according to some studies. While surveying existing works, we found that a majority of works use visible spectrum imaging for detection, while the use of thermal images for this application has been relatively unexplored. Most works also do not focus on model interpretability/explainability or real-time deployment, which hinders their use in practical applications. According to our survey, thermal imaging can prove to be a potential non-destructive approach for this problem, and future works should aim at creating accurate, explainable AI frameworks that can be deployed in real time.