Multimodal Machine Learning For Fruit Ripeness Classification Using Image And Sensor-Based Data Fusion
Dr. P. U. Anitha,
K. Raviteja,
K. Sunhith,
K. Balakrishna,
T. John Sukeerth Reddy
Fruit ripeness detection is a critical task in modern agriculture, directly influencing post-harvest quality,
storage optimization, and market value. Traditional manual inspection methods are often inconsistent,
subjective, and labour-intensive. To overcome these limitations, this project implements a real-time fruit
ripeness classification system using computer vision–based multimodal machine learning, leveraging
YOLO object detection integrated with a live video stream from a laptop camera. The system captures
fruit images through the built-in camera, processes them using the YOLO model, and identifies ripeness
stages based on learned visual patterns such as colour, texture, and surface characteristics. A Stream
lit interface enables seamless real-time prediction, frame annotation, and camera control, making the
solution interactive, efficient, and deployable on consumer-grade hardware. This approach demonstrates
high speed analysis, scalability, and ease of use, presenting a reliable alternative to manual grading.
The system highlights the potential of vision-based machine learning for agricultural automation and
lays the foundation for future multimodal fusion using additional sensors such as temperature, humidity,
and ethylene gas for more precise ripeness evaluation. on centration, and weight—are analysed using a
lightweight MLP/LSTM model for temporal and environmental feature extraction.
