Implementation of Deep Learning Using YOLOv7 and Telegram Notifications for Preventing Illegal Fishing in the Waters of Batam

Abstract

Batam Island is one of Indonesia's outermost islands that directly borders neighboring countries. The implementation of YOLOv7 to detect ships in the waters of Batam is capable of identifying ship objects, with test results after 100 training epochs producing a precision value of 1.00 and a confidence value of 0.882, indicating a high level of confidence in the detection results of the YOLOv7 model. The F1 score of 0.99 at a confidence level of 0.729 shows that this model achieves a high level of accuracy in object detection. Based on the evaluation results using a confusion matrix, it indicates high accuracy for each class in the YOLOv7 model: Ferry 93%, Indonesian Fishing Boat 85%, Malaysian Fishing Boat 89%, Thai Fishing Boat 91%, Vietnamese Fishing Boat 82%, Speedboat 94%, and Tanker 83%. The testing results of the website application integrated with YOLOv7 and Telegram bot produce a website that can detect objects and send notifications, thus expected to prevent illegal fishing.