A Deep Residual Learning Framework using ResNet-152 for Accurate Multi-class Classification of Shrimp Diseases

Abstract

Background: Shrimp aquaculture is a major contributor to global seafood production. Disease outbreaks remain a serious challenge, causing severe economic losses and threatening food security. Timely and accurate disease diagnosis is essential for effective farm management. However, conventional diagnostic methods are labor-intensive, time-consuming and require expert knowledge. Methods: This study proposes a deep learning–based approach for automated multi-class shrimp disease classification using a ResNet-152 convolutional neural network. The publicly available Tiger Shrimp BD dataset was used. It contains 3,574 RGB images of Penaeus monodon grouped into four classes. A substantial portion of the dataset (approximately 72%) consists of synthetically generated images, while the remaining images were collected under real farming conditions across multiple regions of Bangladesh. This combination introduces variability in illumination, background and shrimp orientation but may also affect real-world generalizability. The dataset was divided into training, validation and test sets using a stratified 60:20:20 split to preserve class distribution. Image preprocessing included resizing, normalization and data augmentation. A controlled augmentation pipeline transformed each image using horizontal flip (p=0.5), rotation (-10° to 10°), zoom (0.9-1.1) and contrast adjustment (0.8-1.2), improving robustness and reducing overfitting. Transfer learning was applied and the model was optimized using the Adam optimizer with categorical cross-entropy loss. Result: The proposed ResNet-152 model achieved strong classification performance. An overall test accuracy of 95.82% (95% CI: 93.17%-96.37%) was obtained, along with a Matthews Correlation Coefficient of 0.9442. High classification metrics were observed across all classes. The Healthy and Yellow Head categories showed particularly strong performance. ROC and precision-recall analyses confirmed robust threshold-independent discrimination, with per-class AUC values exceeding 0.98. Qualitative predictions further demonstrated reliable identification of disease patterns under diverse conditions. The proposed model demonstrates strong classification performance. However, reliance on augmented data and lack of external validation may affect generalizability. Future work will focus on independent dataset validation and efficient deployment.