Fish Species Detection Using Deep Learning

Abstract

This research explores the application of deep learning techniques for fish species detection in underwater environments. convolutional neural networks (CNNs) trained on extensive datasets, the study aims to enhance the accuracy and efficiency of species identification. The proposed model demon- strates promising results in differentiating diverse fish species, contributing to advancements in aquatic ecology monitoring and biodiversity conservation. The integration of deep learning in fish species detection holds potential for improving our understanding of underwater ecosystems and supporting sustainable fisheries management. The relative abundance of fish pieces in their habitats on a regular basis and keeping an eye on population fluctuations, this are a crucial task for marine scientists and conservationists diverse automatic computer based fish sample methods have been demonstrated in underwater photos and videos as alternatives to time consuming hand sampling there isn’t however a perfect method for automatically detecting fish and classifying the species this is mostly due to the difficulties in producing clear underwater images and videos which include environmental fluctuations in lightning fish camouflage Dynamic backdrops murky water low resolution shape deformations of moving fish.

Introduction

I. INTRODUCTION

Dive into the future of aquatic research with fish species detection powered by deep learning. fish species identification and detection through the lens of deep learning. In this rapidly evolving field, cutting-edge technologies harness the power of artificial intelligence to revolutionize our ability to discern and classify diverse aquatic life. By delving into the intricacies of deep learning algorithms, we embark on a journey to enhance our understanding of fish biodiversity, contributing to both conservation efforts and the sustainable management of aquatic ecosystems.

We explore the intersection of technology and marine biology, unlocking new possibilities for accurate, efficient, and non-intrusive methods of fish species identification.

Deep convolutional neural network (CNN) models through ensemble learning. This approach aims to significantly enhance diagnostic accuracy and reliability, facilitating early intervention and treatment. Embark on an exploration at the intersection of marine biology and cutting-edge technology as we delve into fish species detection using Convolutional Neural Networks (CNNs). CNNs, renowned for their prowess in image analysis, provide a powerful tool for discerning intricate patterns and features within underwater imagery.

The transformative impact of artificial intelligence, particularly deep learning, on the field of aquatic research. It emphasizes the revolutionary potential of these technologies in discerning and classifying diverse aquatic life. The primary objective is to improve our understanding of fish biodiversity, with a specific focus on fish species identification. This is seen as a crucial contribution to conservation efforts and the sustainable management of aquatic ecosystems. CNNs are recognized for their effectiveness in image analysis, making them a suitable choice for the complex task of fish species identification within underwater imagery.

Ensemble learning combines the predictions of multiple models to enhance diagnostic accuracy and reliability. The goal is to improve the efficiency of fish species identification, enabling early intervention and treatment in the context of aquatic research. The narrative invites readers to explore the intersection of marine biology and cutting-edge technology. It specifically emphasizes how CNNs, known for their ability to analyze images, can be a powerful tool for detecting intricate patterns and features within underwater imagery, facilitating accurate fish species identification.

II. METHODS

A. Data Augmentation

With this technique we generate additional training data by applying various transformations such as rotations, shifts, and flips. This process enhances the diversity of the training dataset, allowing the model to generalize better by being exposed to a wider array of scenarios. By augmenting the data in this manner, the model becomes more robust and capable of performing well on new, unseen data, ultimately improving its overall performance and reliability.

B. Feature Extraction

In the realm of fish species detection, feature extraction entails the process of isolating and analyzing distinctive traits from original data, such as images or sensor readings, which serve as indicators for differentiating between various fish species. These traits encompass a range of characteristics including color patterns, texture, shape, and other morpholog- ical attributes. Commonly employed techniques involve edge detection, analysis of color histograms, and texture assessment, all aimed at distilling pertinent features. The objective is to condense the data’s complexity while retaining its essential information, thereby facilitating subsequent classification or identification tasks through machine learning algorithms. By effectively capturing the unique traits specific to each species, feature extraction plays a pivotal role in automating fish species detection endeavors, thereby contributing to fisheries management, biodiversity evaluation, and ecological research.

C. Deep learning models

In this section, we detail the architectures and configura- tions of the models employed for cross-domain aspect-based sentiment analysis. Two distinctive models, a Long Short-Term Memory (LSTM) network and a Gated Recurrent Unit (GRU), were implemented to capture nuanced sentiment expressions across diverse domains.

1. CNN Models: CNNs are utilized for image classification tasks, including fish species detection, making use of convo- lutional layers to extract features from underwater images. A Convolutional Neural Network (CNN) designed for fish species detection is a sophisticated deep learning architecture specifically tailored to handle the complexities of underwater imagery and the vast diversity of aquatic life. Utilizing its intricate layers and hierarchical feature extraction capabilities, the CNN meticulously dissects images to discern intricate details crucial for species identification. Each convolutional layer acts as a virtual microscope, scanning the input images to capture distinct features such as color patterns, textures, and unique anatomical characteristics like fins and scales. These layers progressively learn to extract increasingly abstract rep- resentations of features, enabling the network to discriminate between subtle differences among various fish species. More- over, pooling layers play a important role in downsampling the feature maps, effectively consolidating essential information while reducing computational complexity. Through extensive training on annotated datasets containing a multitude of fish species, the CNN hones its ability to generalize and make accurate predictions, even with the fluctuations in environmen- tal and Disturbance in image. As a result, this cuttingedge technology not only streamlines the arduous task of species identification but also empowers researchers, conservationists, and fisheries managers with valuable insights for monitoring biodiversity, assessing ecosystem health, and implementing targeted conservation measures in aquatic environments world- wide.
2. MobileNetV2: MobileNetV2 is a lightweight deep learn- ing model optimized for mobile and embedded devices, offer- ing efficient inference with minimal computational resources. MobileNetV2 represents a major progress in the realm of deep learning models, tailored for resource-constrained en- vironments such as mobile devices and embedded systems. MobileNetV2 utilizes an innovative building block known as the inverted residual with a linear bottleneck, which enables the network to capture intricate features while drastically reducing computational costs. By leveraging depthwise separa- ble convolutions, MobileNetV2 efficiently factorizes standard convolutions into separate spatial and channel-wise opera- tions, significantly reducing the number of parameters and computational workload. Furthermore, the network architec- ture incorporates linear bottlenecks that expand the network’s capacity to capture feature representations, enhancing its discriminative power. Through repeated application of these building blocks across multiple layers, MobileNetV2 learns hierarchical representations of visual features essential for accurate speciesidentification. Additionally, techniques such as batch normalization and ReLU6 activation further enhance the network’s stability and non-linearity, ensuring robust perfor- mance across diverse environmental conditions. Trained on large-scale datasets containing annotated fish images, Mo- bileNetV2 fine-tunes its parameters to effectively generalize and make precise predictions, even in challenging underwater environments. As a result, MobileNetV2 stands as a testament to the marriage of cutting-edge deep learning techniques and practical considerations, empowering fishery scientists, conservationists, and environmental managers with a powerful tool for biodiversity monitoring, ecosystem assessment, and sustainable fisheries management.
3. Model performance: The outcomes of fish species identification models have been multifaceted, contributing to significant advancements in environmental monitoring and conservation efforts. With the deployment of advanced tech- nologies and machine learning algorithms, there has been a notable improvement in the accuracy and efficiency of fish species identification, facilitating better management of aquatic ecosystems. This enhanced capability has enabled researchers and conservationists to monitor fish populations more effectively, assess biodiversity trends, and identify areas of concern for targeted conservation interventions. Addition- ally, the availability of live data and predictive analytics generated by these models has empowered policymakers and resource managers to make good well thought decisions re- garding fisheries management and habitat conservation strate- gies, ultimately leading to more sustainable practices and the preservation of aquatic biodiversity. Moreover, efforts to dis- seminate knowledge and raise awareness about the importance of fish conservation have been intensified, with educational programs and public outreach initiatives aimed at engaging communities in conservation efforts. By fostering a greater un- derstanding of the value of aquatic ecosystems and the threats they face, these initiatives have mobilized public support for conservation initiatives and encouraged community involve- ment in monitoring and protecting fish species. Furthermore, the development of user-friendly reporting mechanisms, such as mobile applications and online platforms, has empowered citizens to contribute to conservation efforts by reporting sightings of rare or endangered fish species, illegal fishing activities, or instances of habitat degradation. This decen- tralized approach to data collection has facilitated the rapid detection of environmental threats and the implementation of timely intervention measures, bolstering conservation efforts and safeguarding aquatic biodiversity for future generations.

III. RESULTS

A. Performance Metrics

For assessing the performance of the fish species detection model, several evaluation metrics were employed.

1. Accuracy: Accuracy assesses the overall correctness of the model’s predictions, representing the proportion of accurate predictions to the total instances in the dataset. While accuracy is a fundamental metric, it is important to comple- ment it with other metrics, especially in cases of imbalanced datasets. our model accuracy is 99.94%
2. Precision: Precision measures the accuracy of the model’s positive predictions, indicating the ratio of true posi- tive predictions to the total positive predictions. In the context of fish species detection, a high precision value implies fewer false positives, which is crucial to avoid misidentification of fish species. our model overall precision is 99.99%
3. Recall: Recall, also known as sensitivity or the true positive rate, evaluates the model’s ability to detect all relevant instances within the dataset. In fish species detection, a high recall value signifies the model’s capability to successfully identify all instances of the target fish species.
4. Loss Curves - Train vs. Box loss, Val vs. Box loss: Loss curves provide insights into the training process of the model. ”Train vs. box loss” and ”Val vs. box loss” refer to the loss values during training and validation. The box loss is typically associated with the localization and classification errors made by the model. Examining these curves can help assess how well the model is converging during training and whether overfitting or underfitting is occurring

B. Comparative Analysis

In comparison to established baseline models, our frame- work exhibited significant improvements in both accuracy and F1-score. The comparative analysis underscores the effective- ness of our proposed methodology in enhancing sentiment analysis performance across various domains.

Conclusion

Fish species detection using deep learning represents a promising avenue for revolutionizing the identification process of fish species from images. By harnessing the power of advanced neural network architectures, these systems exhibit the potential to achieve unprecedented levels of accuracy and generalization across diverse datasets. Through the intricate layers of convolutional neural networks (CNNs) and other sophisticated techniques, these models can discern intricate patterns and features from images, which helps us for precise and accurate classification of fish species. The ability to learn from vast amounts of labeled data enables these systems to recognize subtle differences among species, even in instances where human interpretation might falter. Moreover, the scal- ability of deep learning frameworks empowers these systems to handle large datasets efficiently, facilitating the analysis of extensive collections of fish images with remarkable speed and accuracy.

References

[1] Manikandan L. R, Mohamed Rayhan, Prabhu p, Dr. S. R. Preethi ”Fish Recognition and Detection Based on Deep Learning,” in IRJET, vol. 08, pp. 98-102, 2021, e-ISSN: 2395-0056, p-ISSN 2395-0072. [2] M. Bhanumathi, Dr. B. Arthi,”Future Trends and Short- Review on Fish Species Classification Models based on Deep Learning Approaches,” in International Conference on Augmented Intelligence and Sustainable Systems(ICAISS- 2022), IEEE Xplore Part Number: CFP22-CB2ART, pp.1451- 1455, 2022, DOI: 10.1109/ICAISS55157.2022.10011087. [3] S. K. Aruna, N. Deepa and T. Devi, ”Underwater Fish Identification in Real-Time using Convolutional Neural Network,” 2023 7th International Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 2023, pp. 586-591, doi: 10.1109/ICICCS56967.2023.10142531. [4] S. Liawatimena et al., ”A Fish Classification on Images using Transfer Learning and Matlab,” 2018 Indonesian Association for Pattern Recognition International Conference (INAPR), Jakarta, Indonesia, 2018, pp. 108-112, doi: 10.1109/INAPR.2018.8627007 . [5] Aditya Agarwal, Prof. Manonmani S, Gaurav Rawal, Tushar Malani, ”Underwater Fish Detection” In International Journal of Engineering Research Technology (IJERT), Vol. 9, pp. 816-819, April 2020, ISSN: 2278-0181 [6] F. Han, J. Zhu, B. Liu, B. Zhang and F. Xie, ”Fish Shoals Behavior Detection Based on Convolutional Neural Network and Spatiotemporal Information,” in IEEE Access, vol. 8, pp. 126907-126926, 2020, doi: 10.1109/ACCESS.2020.3008698. [7] C. S. Arvind, R. Prajwal, P. N. Bhat, A. Sreedevi and K. N. Prabhudeva, ”Fish Detection and Tracking in Pisciculture Environment using Deep Instance Segmentation,” TENCON 2019 - 2019 IEEE Region 10 Conference (TENCON), Kochi, India, 2019, pp. 778-783, doi: 10.1109/TENCON.2019.8929613. [8] J. Winkler, S. Badri-Hoeher and F. Barkouch, ”Activity Segmentation and Fish Tracking From Sonar Videos by Combining Artifacts Filtering and a Kalman Approach,” in IEEE Access, vol. 11, pp. 96522-96529, 2023, doi: 10.1109/ACCESS.2023.3294710. [9] Kunal Chakraborty, Sourav Maity, Aneesh A. Lotlikar, Alakes Samanta, Jayashree Ghosh, Nagaraja Kumar Masuluri, Naga Swetha and Rose P. Bright, “Modelling of marine ecosystem in regional scale for short term prediction of satellite-aided operational fishery advisories.”, 2019 Journal of Operational Oceanography, pp. S157–S175, doi: 10.1080/1755876- X.2019.1574951.Nhat D. M. Nguyen, Kien N. Huynh, Nhan.N.Vo, Tuan Van Pham,”Fish Detection and Movement Tracking”, International Conference on Advanced Technologies for Communications (ATC) ,2015. [10] A.Rova, G.Mori, L.Dill, ”One fish two butterfish trumpeter: recognizing fish in underwater video”-IAPR Conference on Machine Vision Applications, pp.404-407, May 2007. [11] H. Yao, Q. Duan, D. Li, J. Wang, “An improved K-means clustering algorithm for fish image segmentation”, Math and Com modelling, V58, I3-4, pp790-798, August, 2013. [12] D. Zhang and G. Lu, “Segmentation of Moving Objects in Image Sequence: A Review “, Circuits Systems Signal Processing, Vol. 20, No.2, 2001, Pp. 143-183. [13] H. Celik, A. Hanjalic, E.A. Hendriks, S. Boughorbel, “Online Training of Object Detectors from Unlabeled Surveillance Video”, in Proc. IEEE CVPR, 2008. [14] S. Manen, M. Guillaumin, L. Van Gool, “Prime object proposals with randomized Prim’s algorithm”, ICCV, 2013. [15] Sutton RT, Pincock D, Baumgart DC, Sadowski DC, Fedorak RN, Kroeker KI. An overview of clinical decision support systems: benefits, risks, and strategies for success. NPJ Digit Med. 2020 Feb 6;3:17. doi: 10.1038/s41746-020- 0221-y. PMID: 32047862; PMCID: PMC7005290.

Copyright

Copyright © 2024 Akash Pokharkar, Niranjan Deshmukh , Manmath Biradar. 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.