Sugarcane-PK v1.0: A Scale-Referenced Dataset and MATLAB Implementation for Precision Phenotyping of Pakistani Sugarcane Varieties

Abstract This repository contains the official dataset structure and MATLAB implementation for Sugarcane-PK v1.0, a curated subset of 7,362 scale-referenced RGB images covering five major Pakistani sugarcane varieties. The project facilitates reproducible research in plant phenotyping, precision agriculture, and Edge-AI deployment. 🔬 Project Overview In tropical South Asian contexts, manual varietal identification at sugar mills is prone to error (15–25% misidentification rates), leading to significant economic losses. Sugarcane-PK v1.0 provides a standardized, scale-aware dataset designed to train computer vision models capable of identifying varieties in less than 30 seconds. Key Features 5 Major Varieties: CP77-400, CPF-237, CPF-250, CPF-251, and CPF-253. 8 Morphological Categories: Comprehensive phenotyping including Bud morphology, Internode geometry, Leaf width, Leaf pattern, Cross-section, Bulk cane, Bulk-near distance, and Individual stalks. Scale-Aware Calibration: Every image includes an in-frame metric ruler, enabling precise pixel-to-millimeter calibration for geometric analysis. Real-World Acquisition: Captured using multiple mobile devices across diverse agro-climatic zones in Punjab, Pakistan (Mianwali and Bhakhar) to ensure model robustness. 📂 Repository Structure The repository is organized to support direct integration into deep learning pipelines: . ├── datasets/sugarcane_PK_V1.0/ │ ├── annotations/ # COCO-style JSON annotations │ ├── metadata/ # CSV and Parquet manifests (SHA-256 hashes) │ ├── [FeatureFolders]/ # Images organized by morphology (e.g., bud, leaf) │ └── DATASET_CARD.md # Detailed data specifications ├── code/ # MATLAB source code │ ├── train_resnet50.m # Training script for ResNet-50 │ ├── train_mobilenetv2.m # Training script for MobileNetV2 │ ├── evaluate_model.m # Performance metrics and Confusion Matrix │ └── run_all.m # Master script for reproducibility └── README.md Implementation Details The included MATLAB code provides a baseline for varietal classification using Transfer Learning. It includes scripts for training ResNet-50 and MobileNetV2 architectures, along with evaluation metrics to benchmark performance against the dataset. 3. Keywords (Add these to the "Keywords" section to improve search ranking) Precision Agriculture Plant Phenotyping Computer Vision Sugarcane Deep Learning MATLAB Edge AI Dataset