Detecting and counting harvested fish and measuring fish body lengths in video using deep learning methods
Journal
2019 ASABE Annual International Meeting
Date Issued
2019-01-01
Author(s)
Tseng, Chi Hsuan
Abstract
© 2019 ASABE Annual International Meeting. All rights reserved. The statistics of harvested fish are key indicators for marine resource management and sustainability. In recent years, electronic monitoring systems (EMS) are used to record fishing process. The statistics of harvested fish in the EMS videos are next manually collected and recorded. Manual measurements are, however, time consuming, and labor intensive. The EMS videos usually contain complex background and the illuminations are uncontrolled. This study proposes to automatically detect harvested fish and measure their body lengths in the EMS videos using deep learning. In the study, the fish were detected and segmented from the background at pixel level in the frames of the EMS videos using mask regional-based convolutional neural networks (Mask R-CNN). The counting of fish was determined using distance thresholding. Subsequently, the body length of the fish was next estimated as the distances between the farthest ends of the fish body. The body length of a fish was determined as the mean body length of the fish with top 5 confidence scores predicted by the Mask R-CNN model in the frames that the fish presented. The developed Mask R-CNN model reached a recall of 98.10% and a mean average precision of 94.77% in fish detection. The proposed method for fish counting reached a precision of 73.37% and a recall of 90.12%.
Subjects
Convolutional neural networks | Fish body length | Fish resource management | Instance segmentation | Object detection
SDGs
Other Subjects
Convolution; Fish; Fisheries; Harvesting; Marine biology; Natural resources management; Neural networks; Object detection; Resource allocation; Complex background; Confidence score; Convolutional neural network; Electronic monitoring systems; Fish resources; Learning methods; Manual measurements; Marine resource management; Deep learning
Type
conference paper