Multimodal perishable fruits and vegetables dataset.

Area of Science:

• Agricultural Science
• Computer Science
• Data Science

Background:

• Growing demand for non-invasive methods to assess produce freshness and quality in the agricultural sector.
• Limitations of visual inspection alone in detecting early spoilage indicators.
• Need for advanced technologies to reduce post-harvest losses in export-oriented supply chains.

Purpose of the Study:

• To develop and present a comprehensive multimodal dataset for research in produce quality assessment.
• To support advancements in non-invasive classification, spoilage detection, and shelf-life prediction.
• To facilitate research in multimodal data fusion and deep learning for freshness assessment.

Main Methods:

• Collected IR-Fusion images, sRGB images, and methane concentration readings from six Indian fruits and vegetables (guava, carrot, tomato, Indian gooseberry, banana, mango).
• Allowed specimens to decompose under controlled indoor conditions (natural lighting, ambient temperature, controlled airflow).
• Compiled a dataset of over 14,000 sRGB images, 14,500 IR-fusion images, and 18 methane sensor files, categorized as Normal or Classified (Spoiled/Not_spoiled).

Main Results:

• The dataset integrates thermal, visual, and chemical spoilage indicators for simultaneous analysis.
• It provides a foundation for developing AI-driven, non-invasive quality assessment tools.
• The multimodal design enables more reliable and automated freshness assessment compared to traditional methods.

Conclusions:

• The developed multimodal dataset is a valuable resource for smart agriculture and Agriculture 5.0 initiatives.
• It enables intelligent, automated, and non-invasive produce quality assessment, improving decision-making.
• Supports the reduction of waste and enhancement of food quality monitoring in agricultural and export supply chains.