Identification and Counting of Pirapitinga Piaractus brachypomus Fingerlings Fish Using Machine Learning.

Area of Science:

• Aquaculture
• Computer Vision
• Machine Learning

Background:

• Accurate fish identification and counting are vital for effective aquaculture management, including stocking, harvesting, and marketing.
• Convolutional neural networks (CNNs) are increasingly used for fish counting and identification, with ongoing research into optimizing their learning capabilities.
• Batch normalization is a key technique for enhancing the stability and accuracy of deep learning models.

Purpose of the Study:

• To evaluate the effectiveness of machine learning, specifically CNNs with batch normalization, for identifying and counting pirapitinga (Piaractus brachypomus) fry.
• To determine the optimal batch size for training the CNN model to improve detection and counting accuracy.

Main Methods:

• Utilized one thousand labeled photographic images of pirapitinga fingerlings with bounding boxes for training.
• Developed an adapted CNN model incorporating batch normalization layers within each convolution block.
• Trained the model for 150 epochs, testing batch sizes of 5, 10, and 20.
• Assessed model performance using precision, recall, and mean Average Precision at an Intersection over Union threshold of 0.5 (mAP@0.5).

Main Results:

• Models trained with smaller batch sizes (5 and 10) demonstrated lower performance.
• The model trained with a batch size of 20 achieved superior results, including 96.74% precision, 95.48% recall, 97.08% mAP@0.5, and 98% accuracy.
• Performance metrics indicate a positive correlation between larger batch sizes and improved accuracy in detecting and counting pirapitinga fry.

Conclusions:

• Larger batch sizes significantly enhance the accuracy of machine learning models for detecting and counting pirapitinga fry.
• The findings suggest that optimizing batch size is crucial for developing robust computer vision systems in aquaculture.
• This research provides valuable insights for improving automated fish monitoring systems in farmed environments.