FINDER: Fish Identification using Deep Ensemble Recognition

Classification of Fish Species based on Visual Features

Data Description

Fish-Pak Dataset

1. Fish-Pak: A data set of images of 6 different fish species, i.e., Catla (Thala), Hypophthalmichthys Molitrix (Silver Carp), Labeo Rohita (Rohu), Cirrhinus Mrigala (Mori), Cyprinus Carpio (Common Carp), and Ctenopharyngodon Idella (Grass Carp).

2. Fish-Pak contains 915 high-resolution images captured by Canon EOS 1300D from 3 different places Head Qadirabad, Head Marala and the river of Chenab, Gujrat, Punjab, Pakistan.

3. Fish-Pak data set can be used for the image classification based on their visual features.

4. The data set is referred for the multiclass problem as it holds three visual features, i.e., Scale (color), Body, and Head.

The link to the dataset is attached here.

Croatian Dataset

1. The dataset consists set of images for 12 different fish species, i.e., Chromis Chromis, Coris Julis Female, Coris Julis Male, Diplodus Annularis, Diplodus Vulgaris, Oblada Melanura, Serranus Scriba, Spondyliosoma Cantharus, Spicara Maena, Symphodus Melanocercus, Symphodus Tinca, and Sarpa Salpa.

2. The fishes were recorded with different poses and sizes. The size depends mainly on the distance between fishand camera as well as the size of the fish itself. The source videos were taken under varyinglight conditions.

3. Due to the varying frequency of appearance of each species in ourvideos, the number of images varies between classes. For example, the most frequent species Diplodus Vulgaris has 110 images and the least Sarpa Salpa is presented 17 times.

The link to the dataset is attached here.

Proposed Methodology:

An ensemble of CNN, CNN + Attention and Transformer model to classify fish species based on visual features.

Neural Network

1. DarkNet-53
2. Residual Attention Network-56
3. Vision Transformer (ViT-B16)

Ensemble:

1. Sum Rule
2. Product Rule
3. Majority Voting

Citation:

@article{meshram2025finder,
  title={FINDER: Fish Identification using Deep Ensemble Recognition},
  author={Meshram, Rahul and Kurmi, Ankit and Banerjee, Arnab and Bhattacharjee, Debotosh and Das, Nibaran},
  journal={International Journal of Information Technology},
  pages={1--14},
  year={2025},
  publisher={Springer}
}