A benchmark dataset for primitive Indian paddy field images with deep learning based classification.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Automated crop varietal identification is crucial for precision agriculture.
• Existing datasets may not adequately represent diverse primitive rice varieties.
• Developing robust identification systems requires comprehensive benchmark datasets.

Purpose of the Study:

• To introduce a novel benchmark dataset of primitive Indian paddy field images.
• To establish a deep learning-based classification baseline for automated variety identification.
• To facilitate research in crop monitoring and precision agriculture.

Main Methods:

• A dataset comprising 3400 images of 33 primitive rice varieties was created.
• Images were captured from diverse field locations in Odisha, India.
• A NASNet-Large model was used for feature extraction, followed by an Error Correcting Output Codes (ECOC) classifier.

Main Results:

• The deep learning model achieved high performance metrics, including 0.9424 accuracy and 0.9439 F1-score.
• High specificity (0.9982) and precision (0.9536) were observed.
• The classification pipeline demonstrated robust performance on the benchmark dataset.

Conclusions:

• The introduced dataset and baseline model provide a valuable resource for rice varietal identification research.
• This work contributes to advancing automated crop monitoring and precision agriculture techniques.
• The findings encourage further development of AI-driven solutions in agriculture.