Computer Vision Lab - FishDisease

Flatfish Lesion Detection based on Part Segmentation Approach
and Lesion Image Generation

Seo-Bin Hwang, Han-Young Kim, Chae-Yeon Heo,
Sung Ju Jung, Hieyong Jeong and Yeong-Jun Cho

Chonnam National University

Journal of the World Aquaculture Society 2025

ABSTRACT

The flatfish is a major farmed species consumed globally in large quantities. However, due to the densely populated farming environment, flatfish are susceptible to injuries and diseases, making early disease detection crucial. Traditionally, diseases were detected through visual inspection, but observing large numbers of fish is challenging. Automated approaches based on deep learning technologies have been widely used, to address this problem, but accurate detection remains difficult due to the diversity of the fish and the lack of the fish disease dataset. In this study, augments fish disease images using generative adversarial networks and image harmonization methods. Next, disease detectors are trained separately for three body parts (head, fins, and body) to address individual diseases properly. In addition, a flatfish disease image dataset called FlatIMG is created and verified on the dataset using the proposed methods. A flash salmon disease dataset is also tested to validate the generalizability of the proposed methods. The results achieved 12% higher performance than the baseline framework. This study is the first attempt to create a large-scale flatfish disease image dataset and propose an effective disease detection framework. Automatic disease monitoring could be achieved in farming environments based on the proposed methods and dataset.

OVERVIEW

1. Flatfish Lesion Image Generation (Figure 4–6)

To augment the limited dataset, the study proposes a two-stage lesion synthesis approach:

Lesion Patch Generation: GAN (StyleGAN2-ADA) is trained on real lesion patches to generate realistic variations (Figure 5). Generated patches simulate new lesion types via noise vector manipulation.
Image Harmonization: These patches are blended into non-lesion fish images using foreground segmentation (via SAM) and style-consistent harmonization techniques (Figure 6). Boundary blurring further enhances visual realism. (Result: 4,773 high-fidelity augmented images.)

2. Fish Part Segmentation for Targeted Detection (Figure 7, Figure 10)

Given that lesions vary across body regions, the framework segments the fish into three regions:

Head, Fins, and Body (Figure 10)
PCA-based fish alignment ensures consistent orientation (Figure 8).
Structural cues (tail length, ellipse fitting) are used for part separation without deep models.

Each part is handled by an individual YOLOv8-based detector, enabling fine-grained lesion detection and reducing false positives due to background clutter.