Deep Blue Discoveries: Advancements in Fish Detection and Species Taxonomy Using Deep Learning in the Subaquatic Realm

Abstract

The Deep Learning Based Underwater Fish Detection and Species Classification project employs deep learning methodologies to accurately detect and classify underwater fish species using a dataset consisting of diverse fish images captured from different underwater locations. The project utilizes a Region-based Convolutional Neural Network (R-CNN) to extract image features, ensuring precise species identification. The model\'s performance is evaluated using accuracy metrics to assess its effectiveness in fish detection and species classification. This project holds practical implications for the field of fisheries management and conservation by enabling accurate identification and classification of fish species in underwater environments. Moreover, the project\'s outcomes have the potential to find applications in marine biology and ecology, serving as a valuable tool for studying fish behavior and ecological dynamics within their natural habitats. The project\'s successful implementation underscores the transformative potential of deep learning techniques in the domain of computer vision, specifically in tasks such as image classification, segmentation, and object detection. In conclusion, the Deep Learning Based Underwater Fish Detection and Species Classification project employs advanced deep learning techniques to accurately identify and categorize various fish species in underwater environments. The project\'s diverse dataset, encompassing fish images captured from different underwater locations, enhances its significance for the fields of fisheries management and conservation. The project\'s outcomes have the potential to significantly contribute to our understanding of underwater ecosystems and support endeavors aimed at their protection and preservation.

Introduction

I. INTRODUCTION

Automated techniques, particularly those utilizing deep learning-based approaches, have gained traction in fish detection and species classification in underwater environments due to their significance in scientific, ecological, and conservation applications [2] [3]. While traditional manual methods for fish identification are time-consuming and susceptible to human biases, automated techniques offer a promising solution [2] [3]. Ecosystem monitoring is one of the primary reasons why fish detection and species classification in underwater environments hold such importance. Fish species serve as vital components of aquatic food webs and play crucial roles in maintaining the balance of marine ecosystems [1].

Accurately detecting and classifying fish species enables researchers to assess ecosystem health, dynamics, abundance, distribution, and biodiversity, which are crucial for understanding ecosystem functioning, identifying potential threats, and implementing effective conservation strategies [4] [5]. Additionally, fish detection and species classification contribute significantly to conservation and management efforts. Precise identification of fish species aids in recognizing endangered, threatened, or invasive species, facilitating targeted conservation actions [2]. It also helps in evaluating the impact of human activities, such as climate change, pollution, and fishing, on fish populations and their habitats. By monitoring fish populations and comprehending their dynamics, policymakers and conservationists can develop sustainable fishing practices, establish marine protected areas, and implement measures to mitigate adverse effects on fish habitats [6].

Fish detection and species classification are also essential for biodiversity studies in underwater environments. Fish species richness and diversity serve as vital indicators of overall biodiversity and ecosystem health [1]. Monitoring and cataloging fish species diversity allow researchers to assess the effectiveness of conservation measures, identify areas of high biodiversity, and evaluate ecosystem responses to environmental changes. This information enhances our understanding of ecological interactions and aids in making informed decisions for biodiversity conservation [7].

Furthermore, fish detection and species classification play a crucial role in fisheries management. The fishing industry contributes significantly to global food security and the livelihoods of coastal communities. Accurate identification of fish species is necessary for monitoring catch composition, enforcing fishing regulations, and preventing overfishing or illegal fishing practices [8]. By comprehending fish population dynamics and spatial distribution, fisheries managers can implement sustainable fishing quotas, size restrictions, and gear regulations.

This ensures the long-term sustainability of fish stocks and the preservation of marine ecosystems [9]. Fish detection and species classification also find applications in aquaculture and mariculture. In aquaculture, precise identification of fish species is required to maintain genetic purity and prevent disease outbreaks during controlled breeding and cultivation [8]. In mariculture, which involves the cultivation of marine organisms in ocean environments, fish detection and species classification assist in monitoring stock health, optimizing feed management, and preventing disease transmission. These applications contribute to the sustainable growth of the aquaculture industry while minimizing potential environmental impact [10]. To summarize, fish detection and species classification in underwater environments are crucial for scientific research, ecological monitoring, conservation efforts, fisheries management, and aquaculture practices. Accurate identification of fish species enables a better understanding of ecosystem dynamics, facilitates targeted conservation strategies, and supports sustainable fisheries management. The use of deep learning-based approaches for automated fish detection and species classification holds great promise for advancing our knowledge of marine ecosystems and ensuring their long-term conservation [11] [12] [13] [14] [15].

TABLE I. THE DATASET USED FOR TRAINING

II. METHODOLOGY

A. Dataset Collection and Pre-processing

The underwater image dataset was collected from various sources, ensuring diversity in fish species, underwater environments, and lighting conditions. Permissions were obtained, and ethical considerations were observed. The dataset was pre-processed by removing duplicates and irrelevant images, organizing them into appropriate categories and formats, and ensuring consistency in resolution, color spaces, and file formats. Data augmentation techniques enhanced dataset diversity.

B. R-CNN Model for Feature Extraction

The R-CNN architecture, based on pre-trained models like ResNet accurately extracted features from fish regions. Model adjustments were made for the target classes, and suitable hyperparameters and optimization algorithms were selected. Training progress was monitored, and performance was evaluated using accuracy metrics.

The classification loss measures the difference between predicted and true class probabilities using cross-entropy loss. The regression loss measures the difference between predicted and ground truth bounding box coordinates using smooth L1 loss. The total loss is obtained by summing the classification and regression losses, weighted by a hyperparameter λ.

C. Fish Detection Using R-CNN

Fish detection involved region proposal and feature extraction. Selective search or edge boxes generated fish region proposals, which were filtered based on criteria and fed into the R-CNN model for feature extraction. High-confidence scores determined fish presence, and non-maximum suppression eliminated redundant bounding boxes.

D. ResNet Model for Species Classification

A fine-tuned ResNet model captured intricate features from the R-CNN's extracted feature vectors. The ResNet model was trained using annotated data, associating each feature vector with the corresponding fish species label. Hyperparameters were configured, and the loss function was optimized.

E. YOLO V7 for Fish Detection by Bounding Box

YOLO V7, known for real-time detection, was trained on the annotated dataset. It generated bounding boxes around fish instances, simultaneously predicting class probabilities and box coordinates. The model was fine-tuned, and performance was optimized using suitable hyperparameters and loss functions.

F. System Evaluation and Performance Analysis

The system's performance was evaluated using accuracy, precision, recall, and F1-score metrics on a separate test dataset. Computational efficiency and inference time were analyzed on different hardware platforms.

G. System Deployment and User Interface

The system incorporated a custom user interface developed using Streamlit, HTML, and CSS. Users could upload images or input video streams for real-time fish detection. Detected fish instances were highlighted, and species labels were displayed. Customization options and detailed reports were provided.

H. Ethical Considerations

Ethical guidelines were followed, including obtaining permissions, ensuring privacy and confidentiality, and minimizing harm to fish species. Copyright and intellectual property rights were respected.

I. System Evaluation and Validation

The system underwent rigorous evaluation and validation using diverse datasets. Metrics such as precision, recall, and F1-score quantitatively assessed accuracy. Comparative analysis highlighted improvements over existing methods.

These findings the contribute to the field of underwater fish detection and classification, with implications for marine research and conservation efforts.

ACCURACY PLOT:

III. SOLUTION

This journal presents a comprehensive technical solution using advanced technologies and deep learning models for accurate and efficient underwater fish species detection. The solution includes dataset collection, pre-processing, feature extraction using R-CNN, species classification using ResNet, training, optimization, evaluation metrics, system deployment with Streamlit and HTML/CSS, and considerations for limitations and future work. The proposed solution provides a valuable tool for marine research and conservation efforts. Underwater environments pose unique challenges, and accurate detection is crucial for ecological studies and resource management. The solution leverages deep learning models to address these challenges.

The diverse dataset is collected ethically, pre-processed, and augmented. R-CNN generates fish region proposals and extracts discriminative features, while ResNet enables species classification. The models are fine-tuned using the annotated dataset. The training process optimizes hyperparameters, minimizes the loss function, and evaluates performance using precision, recall, and F1-score. Comparative analysis demonstrates the solution's superiority. Streamlit, HTML, and CSS are used for user-friendly deployment.

Limitations like occlusion and lighting variations need to be addressed, and future work involves expanding the dataset and integrating advanced techniques like YOLO V7 for real-time detection. The solution contributes to marine research and conservation, and further improvements can enhance its applicability in real-world underwater environments.

Overall, the proposed solution contributes to the advancement of marine research and conservation efforts by providing a valuable tool for accurate fish species detection in challenging underwater environments. Further enhancements and refinements hold the potential to broaden its applicability in real-world scenarios.

ACCURACY GRAPH:

IV. RESULTS AND DISCUSSION

A. Fish Detection Performance

The system achieved high accuracy (92.3%) in detecting fish instances within underwater images or videos. Precision (94.8%) and recall (90.5%) rates were also favorable, indicating accurate identification and capturing of fish instances. The F1-score (92.6%) demonstrated robust overall performance in fish detection.

B. Species Classification Performance

The system showed effective species classification with an accuracy rate of 87.6%. This indicates the system's ability to correctly classify fish instances into their respective species categories.

C. Real-Time Fish Detection Performance

The integration of YOLO V7 enabled real-time fish detection with an average FPS of 30 frames, ensuring rapid analysis of underwater video streams.

D. Comparative Analysis

The developed system outperformed existing methods, as evidenced by higher accuracy, precision, recall, and F1-score. This validates the effectiveness of the proposed deep learning-based approach.

E. Discussion of Limitations and Challenges

Challenges include acquiring a diverse and accurate underwater image dataset, generalizing the system to unseen environments and fish species, and addressing occlusions and overlapping instances. Fine-tuning on additional datasets and incorporating advanced algorithms can help overcome these limitations and enhance system performance.

V. FUTURE SCOPE

The future direction of the project entails incorporating video analysis techniques to enable real-time fish species detection and classification in underwater videos. This extension broadens the project's scope, offering opportunities for continuous monitoring, ecological studies, and underwater habitat assessment.

The project's primary objective remains the development of an accurate system for fish species detection and classification, emphasizing efficiency, scalability, and user-friendliness. The potential integration of video analysis further enhances the system's applicability in underwater fish research and conservation endeavors.

Conclusion

By employing deep learning techniques, the project achieved an impressive 95% accuracy in underwater fish species detection and classification. The integration of R-CNN and ResNet models played a pivotal role in accurately localizing fish regions, extracting crucial features, and reliably classifying species, thereby ensuring the system\'s effectiveness. Incorporating YOLO V7 resulted in significant enhancements in both efficiency and accuracy, particularly in fish detection using bounding boxes. The system demonstrated its proficiency in efficiently processing and analyzing underwater images, enabling prompt and accurate fish species detection. This accomplishment has significant implications for underwater research, ecological studies, and conservation efforts, providing a valuable tool for monitoring fish populations and gaining insights into ecosystem dynamics.

References

[1] Kristian Muri Knausgård et al, \"Temperate Fish Detection and Classification: a Deep Learning based Approach,\" IEEE, 2000. [2] Abdullah Albattal, Anjali Narayanan, \"CLASSIFYING FISH BY SPECIES USING CONVOLUTIONAL NEURAL NETWORKS.\" [3] Suxia Cui, Yu Zhou, Yonghui Wang, Lujun Zhai, \"Fish Detection Using Deep Learning,\" Applied Computational Intelligence and Soft Computing, vol. 2020, ID 3738108, 13 pages, 2020. [4] K. He, G. Gkioxari, P. Dollár, and R. Girshick, \"Mask R-CNN,\" 2017 IEEE International Conference on Computer Vision (ICCV), 2017. [5] S. Ren, K. He, R. Girshick, and J. Sun, \"Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks,\" IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017. [6] R. Mandal, R. M. Connolly, T. A. Schlacher, and B. Stantic, \"Assessing fish abundance from underwater video using deep neural networks,\" 2018 International Joint Conference on Neural Networks (IJCNN), 2018. [7] X. Yang et al., \"Instance Segmentation and Classification Method for Plant Leaf Images Based on ISC-MRCNN and APS-DCCNN,\" IEEE Access, vol. 8, pp. 151555-151573, 2020. [8] Ben Saminiano, Arnel Fajardo, Ruji Medina, \"Feeding Behavior Classification of Nile Tilapia (Oreochromis niloticus) using Convolutional Neural Network,\" International Journal of Advanced Trends in Computer Science and Engineering, 2020. [9] Rekha B.S., Srinivasan G.N., Reddy S.K., Kakwani D., Bhattad N. \"Fish Detection and Classification Using Convolutional Neural Networks,\" Computational Vision and Bio-Inspired Computing. ICCVBIC 2019. Advances in Intelligent Systems and Computing, vol 1108. Springer, 2020. [10] Aditya Agarwal, Tushar Malani, Gaurav Rawal, Navjeet Anand, Manonmani S, \"Underwater Fish Detection,\" International Journal of Engineering Research and Technology (IJERT), Volume 09, Issue 04, 2020. [11] Knausgård, K.M., Wiklund, A., Sørdalen, T.K. et al. \"Temperate fish detection, and classification: a deep learning-based approach,\" Applied Intelligence (2021). [12] Vishnu Kandimalla, Matt Richard, “Automated Detection, Classification and Counting of Fish in Fish Passages with Deep Learning” Novel Technologies for Assessing the Environmental and Ecological Impacts of Marine Renewable Energy Systems, 2022. [13] Suja Cherukullapurath Mana and T. Sasipraba, “An Intelligent Deep Learning Enabled Marine Fish Species Detection and Classification Model” International Journal on Artificial Intelligence Tools, 2022 [14] Pooja Prasenan & Chethamangalathu Damodharaprabhu Suriyakala , “Fish species classification using a collaborative technique of firefly algorithm and neural network”, EURASIP Journal on Advances in Signal Processing, 2022. [15] Israa M. Hassoon, “Fish Species Identification Techniques: A Review”, Al-Nahrain Journal of Science, 2022.

Copyright

Copyright © 2023 Swetha ., Sreerambabu , Mohammed Riyaz, Kalidasan . 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.