Fish Image Instance Segmentation Using the Separation of the Mask Produced for Fish Size Measurement

In this article a novel approach for separation of fish image regions implementing instance coefficients and separating masks is proposed. After considering the instance-specific coefficients as well as the separate segmentation for later precise fish size measurements to get the best segmentation accuracy more precisely and reliably. The model was assessed on different types of datasets and showed a good performance and validity in fish dissection. The bounding box detection and the mask segmentation attained the mean Average Precision(mAP)of 91.17% and 96.45%, respectively. The outcome for each IOU threshold had the expected consistency, which proved the model’s resistance to different orientations, backgrounds, and lighting conditions. Visual inspections which were conducted to determine whether the correct size measurements were properly extracted were helpful in validating the fish contour segmentation accuracy. This research reveals the prospects for the next level of segmentation in the marine biology and fishery science area. It is a robust measure in the field of precise fish size measurement from which the development of this field can be achieved in the future.