SoyCotton-Leafs

<b>Overview</b><br>This repository provides a curated image dataset designed for leaf-level detection and segmentation of soybean (<i>Glycine max</i>) and cotton (<i>Gossypium hirsutum</i>) plants under real-world field conditions. Each image is annotated with both bounding boxes and instance segmentation masks, enabling a broad range of computer vision tasks including volunteer plant detection, weed discrimination, and early disease surveillance in precision agriculture.<b>Key Features</b><b>Real Field Images</b>: All 640 images were captured in a commercial farm setting in São Paulo, Brazil, across diverse growth stages and lighting conditions.<b>Dual Annotations</b>: Both <b>bounding boxes</b> (for object detection) and <b>instance-level masks</b> (for pixel-level segmentation) are provided.<b>Multiple Growth Stages</b>: Includes early, mid, and dense canopy phases, ensuring robustness to occlusions and overlapping foliage.<b>High-Quality Labels</b>: Rigorous annotation process with human experts and post-processing (e.g., connected component analysis) to remove extraneous “pixel blobs.”<b>Soybean &amp; Cotton Leaves</b>: Over 12,000 annotated leaves in total—7,221 soybean leaves and 5,190 cotton leaves—enabling comparative studies between the two crops and detection of volunteer plants.<b>Various Use Cases</b>: Ideal for tasks such as selective herbicide application, volunteer crop monitoring, phenotyping, and canopy analysis.<b>Data Structure</b><b>Images</b>: 640 high-resolution RGB images (1600×1200 pixels), split into training, validation, and test subsets, with additional incremental splits for ablation studies (optional).<b>Annotations</b>:Bounding box coordinates in COCO-like JSON files.Instance segmentation masks (either as polygons or bitmasks, depending on the provided format) for each labeled leaf.<b>Metadata</b>: Includes brief details on capture date, environmental conditions, and growth stage (early, mid, or dense canopy).<b>Suggested Applications</b><b>Leaf-Level Detection</b>: Train and evaluate object detectors (e.g., YOLO series, Faster R-CNN) to localize individual cotton or soybean leaves.<b>Semantic/Instance Segmentation</b>: Deploy segmentation models (e.g., Mask R-CNN, DeepLab) to precisely delineate leaf boundaries.<b>Precision Agriculture Analytics</b>: Use detections to quantify volunteer plants, gauge canopy overlap, or assess leaf area index over time.<b>Weed Management</b>: Extend to weed–crop discrimination, particularly useful in rotational systems where volunteer plants often emerge in subsequent seasons.<br>