Source-linked AI summary
Fish Disease Detection Using Image Based Machine Learning Technique in Aquaculture
Md Shoaib Ahmed, Tanjim Taharat Aurpa, Md. Abul Kalam Azad
TL;DR
Fish disease threatens aquaculture, while early salmon disease detection is constrained by limited infrastructure and datasets. The paper introduces image processing with an SVM classifier, evaluates it on a novel salmon dataset, and reports strong classification performance.
Problem
Fish diseases threaten aquaculture, and early salmon disease detection faces limited infrastructure and a lack of existing datasets.
Method
The study combines image preprocessing, segmentation, contrast enhancement, feature extraction, and an SVM classifier to distinguish fresh from infected salmon.
Results
The proposed SVM outperforms Logistic Regression, Decision Tree, and Naive Bayes across the considered performance metrics, with very low misclassification.
Takeaways & Limitations
The work provides a machine-learning-based salmon disease classification system using a novel real-world dataset and image-processing pipeline.
Takeaways & Limitations
Future work is needed across different fish datasets, with larger datasets, CNN architectures, and real-life IoT implementation.
Abstract
from arXiv · showhide
Fish diseases in aquaculture constitute a significant hazard to nutriment security. Identification of infected fishes in aquaculture remains challenging to find out at the early stage due to the dearth of necessary infrastructure. The identification of infected fish timely is an obligatory step to thwart from spreading disease. In this work, we want to find out the salmon fish disease in aquaculture, as salmon aquaculture is the fastest-growing food production system globally, accounting for 70 percent (2.5 million tons) of the market. In the alliance of flawless image processing and machine learning mechanism, we identify the infected fishes caused by the various pathogen. This work divides into two portions. In the rudimentary portion, image pre-processing and segmentation have been applied to reduce noise and exaggerate the image, respectively. In the second portion, we extract the involved features to classify the diseases with the help of the Support Vector Machine (SVM) algorithm of machine learning with a kernel function. The processed images of the first portion have passed through this (SVM) model. Then we harmonize a comprehensive experiment with the proposed combination of techniques on the salmon fish image dataset used to examine the fish disease. We have conveyed this work on a novel dataset compromising with and without image augmentation. The results have bought a judgment of our applied SVM performs notably with 91.42 and 94.12 percent of accuracy, respectively, with and without augmentation.
1. Introduction
Aquaculture supports food and commercial production but faces environmental and disease-related challenges, especially in salmon farming. The paper proposes an SVM-based image-classification framework supported by image processing and evaluates it with and without augmentation.
- Aquaculture context: Aquaculture cultivates aquatic organisms under controlled freshwater or saltwater conditions for food and commercial products.Mariculture and fish farming are identified as major forms of aquaculture.
- Aquaculture context: Large-scale aquaculture can threaten marine and coastal environments through habitat degradation, nutrient impacts, and waste discharge.
- Motivation: 53% of all fish and invertebrate production was attributed to global aquaculture as of 2020, while farmed salmon production exceeded 2.6 million tonnes in 2019.
- Motivation: Fish diseases threaten aquaculture because dense fish populations and stressful confined conditions can favor contagious disease development and spread.
- Proposed approach: The proposed framework extracts image features through processing and classifies infected versus fresh salmon using a support vector machine.
- Evaluation: The study evaluates the SVM framework with and without image augmentation and compares it with other classification models using evaluation metrics.
2. Related Work
Prior fish-disease studies used image processing, feature extraction, and machine learning, but the paper identifies limited depth in salmon disease classification for aquaculture. It positions the proposed work as addressing this gap with an SVM-based approach.
- Image-processing approaches: Existing fish-disease detection work used image segmentation and feature descriptors such as HOG and FAST, but did not specify the exact feature-extraction technique.
- Combined approaches: A combination of FAST, PCA, and a neural network achieved 86 percent accuracy for detecting Epizootic Ulcerative Syndrome.
- SVM-related approaches: Other research applied morphological operations and segmentation to define regions of interest before SVM classification, including under low-resolution and visually similar conditions.
- Research gap: The paper concludes that prior studies lacked in-depth salmon fish disease classification focused on aquaculture.
3. Preliminary and Proposed Framework
The proposed framework standardizes and enhances fish images, segments infected regions, extracts feature vectors, and classifies infection with SVM. It uses image-processing stages before supervised binary classification.
- Image preprocessing: Adaptive histogram equalization enhances contrast to compensate for information loss caused by image magnification and resizing.The implementation uses CLAHE, whose clipping level controls contrast enhancement and histogram noise.
- Segmentation: RGB images are converted to L*a*b color space, where k-means clustering partitions images into regions and separates infected areas from fish images.The authors state that k-means segments images efficiently in L*a*b space rather than RGB space.
- Feature extraction: Co-occurrence and statistical feature vectors are acquired from infected areas for subsequent classification.The extracted vectors are supplied to SVM as the representation of segmented disease regions.
- Classification: SVM constructs a maximum-margin hyperplane to partition feature vectors into infection classes using a kernel function.The framework describes support vectors as points influencing hyperplane positioning and identifies linear kernels among possible kernel choices.
- System architecture: The system architecture has building and deployment phases, with labeled images processed as training data during building.The image-processing sequence includes cubic splines interpolation, adaptive histogram equalization, and RGB-to-L*a*b conversion.
4. Evaluation
The evaluation uses a novel salmon-fish dataset, image preprocessing and feature extraction, and SVM classification assessed with multiple performance metrics. Experiments compare datasets with and without augmentation while examining statistical and GLCM features.
- Dataset: The dataset contains 266 salmon-fish images, split into 231 training and 35 testing images without augmentation.
- Dataset: Image augmentation expands the dataset to 1,105 training images and 221 testing images.
- Feature extraction: The evaluation extracts statistical and GLCM features, including contrast, correlation, energy, entropy, and homogeneity.
- Classification: A linear SVM classifies ten-dimensional feature vectors containing five statistical and five GLCM features.
- Performance metrics: The study evaluates predictions with confusion-matrix measures, accuracy, precision, recall, specificity, F1 score, FPR, FNR, TPR, and ROC-AUC.
5. Experimental Results
The experiments evaluate the proposed SVM for fresh-versus-infected salmon classification on regular and augmented data, using class-wise metrics, confusion matrices, ROC curves, and classifier comparisons. Overall accuracy reaches 91.42% without augmentation and 94.12% with augmentation.
- The SVM classifies fresh and infected fish using class-wise accuracy, sensitivity, precision, F1 score, and specificity.With augmentation, fresh-fish sensitivity is 98.46% and infected-fish recall is 98.13%; without augmentation, fresh- and infected-fish F1 scores are 96.23% and 97.08%.
- The infected-fish class has higher accuracy than the fresh-fish class in both evaluation settings: 93.50% and 94.90%, respectively.
- The confusion matrices show two fresh fish misclassified as infected and one infected fish misclassified as fresh in one setting, versus seven and six errors in the other.
- 94.12% accuracy is obtained with image augmentation, compared with 91.42% without augmentation.
- ROC analysis reports micro-average AUC scores of 96.20% without augmentation and 98.12% with augmentation, with macro-average scores of 95.93% and 96.71%.
- With augmentation, SVM evaluation metrics outperform decision tree, logistic regression, and naïve Bayes in the reported comparison.Decision tree accuracy is 81.54%, while logistic regression accuracy is 80.0%; its remaining reported metrics are below SVM and above the other compared classifiers.
6. Discussion
The discussion frames salmon disease detection as an underdeveloped area lacking an existing dataset and explains the proposed image-processing and SVM pipeline. The reported results support the selected classifier and the paper’s novel salmon-focused dataset.
- The study introduces a novel salmon fish disease dataset because no existing dataset was available for this research purpose.The dataset contains images of fresh and infected fish used as classifier inputs.
- SVM is selected to classify infected and fresh salmon fish, and the paper reports higher accuracy, precision, sensitivity, specificity, and F1 score than three alternative classifiers.The alternatives are logistic regression, decision tree, and naïve Bayes.
- Image preprocessing applies cubic spline interpolation, adaptive histogram equalization, and k-means segmentation before classification.These stages resize, enhance contrast, and segment the input images before feature extraction and SVM classification.
- The classifier evaluation uses multiple metrics and confusion-matrix visualization, with the reported number of misclassifications described as very low.
- The ROC curve evaluates SVM performance across classification thresholds by plotting true positive rate against false positive rate.
- Prior fish-disease studies rarely focus explicitly on salmon, and some use only image processing rather than an intelligent classification system.
7. Conclusion and Future Work
The work develops an SVM-based fish disease classifier using a novel real-world dataset and updated image-processing techniques. Future work targets CNNs, IoT deployment, broader datasets, and increased salmon-image coverage.
- Conclusion: The study trains an SVM classifier to distinguish fresh and infected fish using real-world datasets with and without augmentation.The datasets contain 163 infected and 68 fresh images without augmentation, and 785 infected and 320 fresh images with augmentation.
- Conclusion: Updated preprocessing uses k-means segmentation, cubic spline interpolation, and adaptive histogram equalization to make input images more adaptable to classification.
- Conclusion: The proposed classifier is reported as outperforming three comparison models on the study’s real-world novel dataset.
- Future Work: Future work will investigate CNN architectures, real-life IoT implementation, additional fish datasets, and a larger salmon-image dataset.The authors link these directions to broader usability and more precise disease identification.