Deep Learning Based Fish Species and Freshness Detection Using Convolutional Neural Networks

Abstract

Maintaining fish freshness is essential in the seafood industry to ensure product safety, quality, and consumer satisfaction. Conventional assessment techniques primarily depend on manual inspection or laboratory-based analysis, which may be subjective, time-intensive, and unsuitable for rapid evaluation. This study introduces a deep learning–driven method for identifying fish species and determining freshness levels using Convolutional Neural Networks (CNN). The framework categorizes fish images into defined freshness classes, including Fresh, Medium, and Spoiled. A MobileNet-based model is adopted due to its lightweight architecture and efficient computational performance. The dataset comprises annotated fish images collected under different storage conditions. Model training and evaluation are conducted using standard classification metrics. Experimental findings indicate strong predictive accuracy and demonstrate the feasibility of deploying the system for real-time applications in seafood markets and processing facilities.

Introduction

Fish is a highly nutritious and widely consumed food source, but it is extremely perishable due to rapid microbial and biochemical spoilage after harvest. Ensuring fish freshness is essential for food safety, quality control, and reducing economic losses. Traditional freshness evaluation methods rely on sensory inspection (appearance, smell, and texture), which are subjective, inconsistent, and require expert judgment.

With advancements in machine learning, IoT, and computer vision, automated freshness detection systems have been developed. Previous research has explored IoT-based monitoring, deep learning models, electronic noses, gas sensors, and hardware-based detection techniques. However, many existing solutions depend on expensive sensors or laboratory equipment, limiting their use in small-scale or real-world market environments.

To address these limitations, this study proposes an intelligent fish species and freshness detection system based on the lightweight MobileNetV2 deep learning architecture. The system performs image acquisition, preprocessing, and classification to identify both fish species and freshness levels (Fresh, Medium, Spoiled). It formulates the problem as a 16-class multi-class classification task, enabling simultaneous prediction of species and freshness in a single model.

The dataset contains 1,959 fish images across six species and three freshness categories, divided into training and validation sets. The system is implemented on an edge device (Raspberry Pi), where images are captured using a camera module, preprocessed (resizing to 224×224, normalization), and processed locally using MobileNetV2.

MobileNetV2 is chosen for its efficiency, using depthwise separable convolutions to reduce computational complexity. The final output provides real-time classification results, making the system suitable for deployment in fish markets and seafood processing environments. Additionally, results can be converted into Tamil speech output, improving accessibility for regional users.

Conclusion

This study presented a deep learning-based system for fish freshness detection using the MobileNet convolutional neural network. The model was trained on a dataset of fish images and successfully classified them into different freshness categories. Experimental results showed that the proposed system achieved high classification accuracy, with training accuracy close to 99% and validation accuracy around 95.87%. The lightweight structure of MobileNet makes the model efficient and suitable for real-time applications. The proposed approach provides a practical and automated solution for fish freshness assessment, which can support food quality monitoring in fish markets and seafood industries.

References

[1] Bhuiyan M. A., Haider S. A. R., Haque A., Hasan M., and Rejwanuddin M., “IoT-Based Meat Freshness Classification Using Deep Learning,” 2024. [2] Chang S., Kim H., and Lee D., “A Fish Meat Freshness Detector for IoT-Based Seafood Market Applications,” 2024. [3] Chen M., and Huang Y., “Composited FishNet: Fish Detection and Species Recognition from Low-Quality Underwater Videos,” 2023. [4] Chen Y., Li Q., and Zhang H., “A CNN-Based E-Nose Using Time-Series Features for Food Freshness Classification,” 2023. [5] Choi K., Lee S., and Park J., “A CMOS Fish Spoilage Detector for IoT Applications of Fish Markets,” 2023. [6] Chung W.-Y., Mai N.-D., and Xu K., “Self-Powered Meat Freshness Monitoring with GAN-Optimized 3D Antenna and Ensemble Learning,” 2024. [7] Kumar P., and Tan S., “Fish Detection Using Electrical Impedance Spectroscopy,” 2023. [8] Rahman A., et al., “A Computer Vision-Based Behavioral Study and Fish Counting in a Controlled Environment,” 2022. [9] Wang L., Xu J., and Zhao F., “A Portable Real-Time Electronic Nose for Evaluating Seafood Freshness Using Machine Learning,” 2023.

Copyright

Copyright © 2026 Arul Selvan G, Aarthi M, Kaviya K, Radhi Devi K R. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.