Model Architecture Comparison for Rice Leaf Disease Detection

Pakistan is the fourth-largest rice producer and exporter globally. This dataset contains a comparative evaluation of five deep transfer learning architectures (InceptionV3, DenseNet201, ResNet152V2, EfficientNetV2L, MobileNetV2) trained on a balanced set of 1914 rice leaf image samples. The work, authored by Ummer Shakeel and last updated in May 2026, applies post-hoc visualization techniques like GrabCut segmentation to enhance interpretability.

Use Cases

• Benchmarking deep learning model performance for agricultural image classification based on the described comparative evaluation.
• Developing explainable AI systems for plant pathology using the post-hoc visualization techniques mentioned.
• Training lightweight models for edge deployment in agriculture, informed by the MobileNetV2 results described.
• Reproducing research on transfer learning for crop disease detection using the publicly available dataset referenced.

Strengths

• Dataset is licensed under CC-BY-4.0, permitting open sharing and adaptation.
• Comparative results for five distinct model architectures are provided, including specific accuracy rates.
• Methodology is described in detail, including the use of 1914 image samples and training over 30 epochs.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset file is very small (5.5 KB), suggesting it contains summary results, not the primary image data.

Provenance

Source

figshare

Collection Method

Images were systematically processed using Python with TensorFlow and a GPU, according to the description.

Time Range

null

Freshness

Last updated 2026-05-07 17:40:57; freshness should be verified.

Geography

Research context focuses on Pakistan's rice cultivation, but specific image collection geography is not stated.