Computational Complexity of Evaluated Models for Tea Disease Detection

2,000 annotated images of tea leaves form the basis for evaluating deep learning models for disease detection. The dataset was created by Sakibul Hasan Chowdhury and last updated in May 2026. It contains performance metrics for seven neural network architectures, including a proposed hybrid model that achieved 96.80% accuracy.

Use Cases

• Benchmarking image classification models based on the reported accuracy, AUC, and convergence stability metrics.
• Developing disease detection systems for tea crops based on the described classes of Blight, Red Rust, Helopeltis, and Healthy.
• Researching hybrid neural network architectures based on the described fusion of EfficientNetV2 and MobileNetV3 feature extractors.
• Evaluating model suitability for IoT edge devices in agriculture based on the computational balance discussed in the description.

Strengths

• Dataset is based on 2,000 real-world annotated images, evenly distributed across four target classes.
• Model evaluation includes specific performance metrics: a peak accuracy of 96.80% and a macro AUC of 0.9980 for the proposed model.
• Provides a direct comparison of seven architectures, including Vision Transformer and several CNNs, with their specific accuracies listed.

Limitations

• Row count is unknown, which may limit suitability assessment for large-scale training.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is very small at 5.5 KB, indicating it likely contains summary results rather than raw image data.

Provenance

Source

Sakibul Hasan Chowdhury via figshare

Collection Method

Likely derived from training and evaluating deep learning models on a proprietary image dataset.

Freshness

Last updated 2026-05-15 17:55:10; freshness should be verified.

Geography

Research context involves tea cultivation in Bangladesh.