Łącki Mirosław
Faculty of Navigation, Gdynia Maritime University, 81-225 Gdynia, Poland.
Sensors (Basel). 2024 Nov 25;24(23):7505. doi: 10.3390/s24237505.
The article describes the use of deep neural networks to detect small floating objects located in a vessel's path. The research aimed to evaluate the performance of deep neural networks by classifying sea surface images and assigning the level of threat resulting from the detection of objects floating on the water, such as fishing nets, plastic debris, or buoys. Such a solution could function as a decision support system capable of detecting and informing the watch officer or helmsman about possible threats and reducing the risk of overlooking them at a critical moment. Several neural network structures were compared to find the most efficient solution, taking into account the speed and efficiency of network training and its performance during testing. Additional time measurements have been made to test the real-time capabilities of the system. The research results confirm that it is possible to create a practical lightweight detection system with convolutional neural networks that calculates safety level in real time.
本文描述了利用深度神经网络检测位于船只航道上的小型漂浮物体。该研究旨在通过对海面图像进行分类，并对检测到的漂浮在水面上的物体（如渔网、塑料碎片或浮标）所造成的威胁程度进行评估，来评价深度神经网络的性能。这样一种解决方案可以作为一个决策支持系统，能够检测并告知值班驾驶员或舵手可能存在的威胁，并降低在关键时刻忽略这些威胁的风险。考虑到网络训练的速度和效率及其在测试期间的性能，对几种神经网络结构进行了比较，以找到最有效的解决方案。还进行了额外的时间测量，以测试该系统的实时能力。研究结果证实，利用卷积神经网络创建一个能够实时计算安全水平的实用轻量级检测系统是可行的。
