Scholarlink

Fish species identification is crucial to the fisheries sector, ensuring compliance and consumer safety. This paper presents a hybrid approach utilising multiple deep learning models to address these challenges effectively. We developed five models specifically designed to classify India’s most common fish species: Pomfret, Mackerel, Black Snapper, Indian Carp, Prawn, Pink Perch, Black Pomfret, and Indian Flathead. The accuracy of the Convolutional Neural Network (CNN) model reached 96.16%, thanks to precise measurements of colour, body shape, and fin configuration. Additionally, the inclusion of residual connections in ResNet50 enhanced feature learning, resulting in an accuracy of 86.4%. VGG16 and VGG19 achieved their highest accuracies at 96.83% and 96.99%, respectively. This multi-model approach effectively supports species classification, creating a comprehensive system to ensure quality and sustainability in fisheries. It also improves sorting and grading processes, making it easier for fish to comply with species-specific regulations, thereby enhancing efficiency in the fish market.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Novel Hybrid Architecture for Classification of Indian Fish Species Using Deep Learning Techniques

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Rithesh Raj,
  • V. Francis Densil Raj,
  • P. Daksh Shanthram

摘要

Fish species identification is crucial to the fisheries sector, ensuring compliance and consumer safety. This paper presents a hybrid approach utilising multiple deep learning models to address these challenges effectively. We developed five models specifically designed to classify India’s most common fish species: Pomfret, Mackerel, Black Snapper, Indian Carp, Prawn, Pink Perch, Black Pomfret, and Indian Flathead. The accuracy of the Convolutional Neural Network (CNN) model reached 96.16%, thanks to precise measurements of colour, body shape, and fin configuration. Additionally, the inclusion of residual connections in ResNet50 enhanced feature learning, resulting in an accuracy of 86.4%. VGG16 and VGG19 achieved their highest accuracies at 96.83% and 96.99%, respectively. This multi-model approach effectively supports species classification, creating a comprehensive system to ensure quality and sustainability in fisheries. It also improves sorting and grading processes, making it easier for fish to comply with species-specific regulations, thereby enhancing efficiency in the fish market.