Coffee and Black Gram Leaf Damage Segmentation Model Results

A PDF document details a computer vision model for segmenting leaf lesions on coffee and black gram plants. The model, integrating DinoV3 with a Unet framework, achieved an Intersection over Union of 78.31% and Pixel Accuracy of 88.00% on the coffee dataset, reducing inference time by 93.6%.

Use Cases

• Benchmark leaf lesion segmentation performance using the reported 78.31% IoU and 88.00% Pixel Accuracy metrics on coffee leaf images.
• Analyze computational efficiency gains from the model architecture, which reduced inference time from 63.41s to 4.07s.
• Validate model scalability and cross-domain reliability using results from the coffee, black gram, and larger AMG_HS datasets mentioned.

Strengths

• Model performance is quantitatively validated, achieving a 78.31% IoU and 88.00% Pixel Accuracy on the coffee leaf dataset.
• Demonstrates significant computational efficiency, reducing inference time by approximately 93.6%.
• Experiments validate robustness across multiple datasets, including coffee, black gram, and AMG_HS.

Limitations

• The dataset consists of a single 395.1 KB PDF file, providing only textual and numerical results, not the underlying image data.
• Experimental scope is limited to specific crops (coffee, black gram) and one larger benchmark (AMG_HS), limiting generalizability claims.
• The document format restricts direct analysis; raw image data, labels, and model code are not included.

Provenance

Source

figshare, authored by Wei Chen.

Collection Method

Research results from experiments applying a DinoV3-Unet integration model to agricultural image datasets.

Time Range

null

Freshness

null

Geography

null