[Paper Review] PlantSeg: A Large-Scale In-the-wild Dataset for Plant Disease Segmentation

Tianqi Wei, Zhi Chen+4 more|arXiv (Cornell University)|Sep 6, 2024

Smart Agriculture and AI8 citations

TL;DR

PlantSeg introduces a large-scale in-the-wild dataset with pixel-level disease segmentation masks for 115 diseases across 34 plants, enabling benchmarking of plant disease segmentation in realistic conditions. It benchmarks multiple baselines and shows strong performance of SegNeXt on this dataset.

ABSTRACT

Plant diseases pose significant threats to agriculture. It necessitates proper diagnosis and effective treatment to safeguard crop yields. To automate the diagnosis process, image segmentation is usually adopted for precisely identifying diseased regions, thereby advancing precision agriculture. Developing robust image segmentation models for plant diseases demands high-quality annotations across numerous images. However, existing plant disease datasets typically lack segmentation labels and are often confined to controlled laboratory settings, which do not adequately reflect the complexity of natural environments. Motivated by this fact, we established PlantSeg, a large-scale segmentation dataset for plant diseases. PlantSeg distinguishes itself from existing datasets in three key aspects. (1) Annotation type: Unlike the majority of existing datasets that only contain class labels or bounding boxes, each image in PlantSeg includes detailed and high-quality segmentation masks, associated with plant types and disease names. (2) Image source: Unlike typical datasets that contain images from laboratory settings, PlantSeg primarily comprises in-the-wild plant disease images. This choice enhances the practical applicability, as the trained models can be applied for integrated disease management. (3) Scale: PlantSeg is extensive, featuring 11,400 images with disease segmentation masks and an additional 8,000 healthy plant images categorized by plant type. Extensive technical experiments validate the high quality of PlantSeg's annotations. This dataset not only allows researchers to evaluate their image classification methods but also provides a critical foundation for developing and benchmarking advanced plant disease segmentation algorithms.

Motivation & Objective

• Highlight the need for high-quality, in-the-wild plant disease segmentation data for robust models.
• Describe the PlantSeg dataset with 11,458 images and 115 disease categories across 34 plants, including detailed segmentation masks.
• Establish a baseline performance on PlantSeg using representative segmentation models and analyze annotation quality and dataset characteristics.

Proposed method

• Dataset construction from web-sourced images using keyword-driven collection across 34 plants and 115 diseases.
• Rigorous data cleaning with cross-annotator validation and expert pathologist review.
• Detailed segmentation annotation using polygon masks with LabelMe, including handling of overlapping lesions and deformities.
• Metadata curation including plant, disease, resolution, mask ratio, and train/test split information.
• Benchmarking with four semantic segmentation models (SAN, DeepLabv3, DeepLabv3+, SegNeXt) and various backbones.
• Evaluation using MIoU and mean accuracy (mAcc).

Experimental results

Research questions

• RQ1 Can a large-scale, in-the-wild plant disease segmentation dataset improve the generalization of segmentation models to field conditions?
• RQ2 How do state-of-the-art semantic segmentation methods perform on complex, real-world plant disease masks compared to laboratory datasets?
• RQ3 What insights do segmentation mask ratios and image resolutions provide for model design and evaluation in plant pathology?
• RQ4 Does backbone size or model architecture significantly influence performance on in-the-wild plant disease segmentation?
• RQ5 Can the dataset serve as a comprehensive benchmark for developing plant disease segmentation methods?

Key findings

• PlantSeg contains 11,458 images with 115 disease categories across 34 plants, plus 8,000 healthy plant images for plant-type categorization.
• Annotation quality was validated with expert pathologist review and cross-annotator consistency checks.
• SegNeXt with MSCAN-L backbone achieves the highest MIoU (44.52%) and mAcc (59.95%).
• SAN with ViT-L/14 backbone outperforms ViT-B/16, achieving MIoU 36.91% and mAcc 52.81%.
• DeepLabv3 and DeepLabv3+ improve with deeper backbones (ResNet-101) but lag behind SegNeXt on PlantSeg.
• Baseline results show substantial gains by using larger backbones and multi-scale convolutional approaches on real-world disease masks.