Development of Fishing Vessel Identification Model Based on Deep Neural Network

Abstract

This paper presents a deep neural network (DNN)‐based model to recognize fishing vessels. In Taiwan, the vast majority of small fishing vessels are not equipped with an automatic identification system (AIS). As a consequence, the staff in a fishing port administration become heavily loaded when monitoring and managing the fishing vessels accessing a port. The workload is expected to be eased using this work. For the first time in the literature, a captured fishing vessel image was converted to a 128‐dimensional embedding for recognition purposes. The presented model gave a false positive rate (FPR) as low as 1.13% and an accuracy up to 99.47% at threshold = 0.772379. Finally, all the performance metrics, namely, the true positive rate (TPR), the FPR, precision and accuracy, are actually functions of the threshold which can be specified by users to meet specific requirements. © 2022 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC.

Approaches to Upgrading the Performance of Fishing Vessel Recognition Technology

“…First, Table 1 gives a comparison between the datasets used in the original proposal and this work. As referenced previously, the number of recognizable fishing vessels was significantly reduced from 156 in the original proposal (5) to 272 in this work. Moreover, the numbers of images of different vessels were made as uniform as possible for a higher generalization ability, and accordingly, the standard deviation (STD) of the number of images per fishing vessel was considerably reduced from 62.00 to 16.93 in this work for improved model training.…”

“…An experiment in this work has two stages and was carried out in exactly the same way as in the original proposal. (5) In Stage 1, the optimal threshold was determined by optimizing the overall performance of the presented model, conducted on the training set in Table 2, with respect to the threshold, as in the Labeled Faces in the Wild database. (10) In Stage 2, a performance test was conducted on the test set in Table 2 to obtain the performance metrics: true positive rate (TPR), false positive rate (FPR), precision, and accuracy.…”

“…As opposed to techniques to recognize hull identification numbers (HINs), (1,2) a face recognition technique (3,4) was used for fishing vessel recognition (5) for the first time. The novel recognition technique in Ref.…”

“…The novel recognition technique in Ref. 5 was found to outperform HIN counterparts mainly owing to the fact that most poorly maintained fishing vessels are pictured with an incomplete HIN, which led to misidentification cases.…”

“…This paper is actually an extended work of the original proposal, (5) and there are two steps to improve the performance of fishing vessel recognition. As will be detailed later, the distribution of images in a dataset was modified, the recognition model in the original proposal was…”