Object recognition on patrol ship using image processing and convolutional neural network (CNN)

Abstract

Autonomous patrol vessels are state border patrol vessels equipped with cameras and image processing capabilities to detect objects around them. This prototype of ship can recognize a detected object; it used an image classification method called Convolutional Neural Network (CNN). So, it will minimize the occurrence of accidents on patrol boats. Input image in the form of RGB will experience feature extraction using a convolution layer. In the classification layer there is an artificial neural network with backpropagation to classify objects against a predetermined dataset. The detection value of the obtained vessel is operated by a predetermined actuator. In the final classification results the object recognition in the form of ships have a quite high accuracy. The average accuracy value is 96.59 percent with a sufficient light condition and RGB image input is taken in real-time.