Effects of image data quality on a convolutional neural network trained in-tank fish detection model for recirculating aquaculture systems

Authors: RANJAN, RAKESH - Freshwater Institute; SHARRER, KATA - Freshwater Institute; TSUKUDA, SCOTT - Freshwater Institute; GOOD, CHRISTOPHER - Freshwater Institute

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/8/2023
Publication Date: 1/12/2023
Citation: Ranjan, R., Sharrer, K., Tsukuda, S., Good, C. 2023. Effects of image data quality on a convolutional neural network trained in-tank fish detection model for recirculating aquaculture systems. Computers and Electronics in Agriculture. 205:107644. https://doi.org/10.1016/j.compag.2023.107644.
DOI: https://doi.org/10.1016/j.compag.2023.107644

Interpretive Summary: Individual fish detection in large, high-density recirculating aquaculture system tanks is a crucial and challenging problem. Artificial intelligence approaches can allow for non-invasive fish detection to monitor fish behavior, biomass, health, and feed conversion ratio, and thus answer fish production-related questions and assist growers with important management decisions. This study was conducted to develop a computer vision tool for fish detection and investigate the effects of sensor selection, image quality, data size, imaging conditions, and pre-processing operations on detection accuracy. An optimized fish-detection model was developed that effectively detected whole and partial fish in the test image with satisfactory performance. This information can be used to investigate the effect of various physical and biological stresses on fish behavior. Real-time monitoring of fish activity during a feeding event can also be utilized to optimize fish growth and FCR and to minimize feed waste through automated adjustment of the feed delivery rate.

Technical Abstract: Artificial intelligence can answer fish production-related questions and assist growers with important management decisions in recirculating aquaculture systems (RAS). However, convolutional neural network-aided machine learning (ML) approaches are data-intensive, with model accuracy subject to the input image quality. Underwater imagery data acquisition, relatively high fish density, and water turbidity impart major challenges in acquiring high-quality imagery data. This study was conducted to investigate the effects of sensor selection, image quality, data size, imaging conditions, and preprocessing operations on the ML model accuracy for fish detection under RAS production conditions. An imaging platform (RASense1.0) was developed with four off-the-shelf sensors customized for underwater image acquisition. Data acquired from the imaging sensors under two light conditions (i.e., ambient and supplemented) were arranged in sets of 100 images and annotated as partial and whole fish. The annotated images were augmented and trained using a custom YOLOv5 ML model. There was a substantial improvement in mean average precision (mAP) and F1 score while increasing the size of the image datasets up to 700 images and 80 epochs. Similarly, image augmentation substantially improved model accuracy for smaller dataset models trained with less than 700 images. Beyond this, there was no improvement in mAP (~85%). Sensor selection significantly affected model precision, recall, and mAP; however, light condition did not demonstrate considerable effect on model accuracy.