The preliminary study of artificial intelligence based on convolutional neural network as a corrosion detection tool on ship structures

Abstract

Technological advances and developments in the Internet of Things (IoT) have made many people and companies aware of using artificial intelligence as a tool to speed up work processes. Deep learning which is a part of artificial intelligence is an important application in the application of Convolutional Neural Network (CNN) for image classification and detection. Convolutional Neural Network (CNN) is an innovation in the development of Multilayer Perceptron (MLP) in image processing. This research aims to conduct a preliminary study on the application of the Convolutional Neural Network (CNN) to obtain a corrosion classification based on the severity of the area on the ship's structure and the appropriate Convolutional Neural Network (CNN) architecture to detect and classify corrosion based on the detection error value. The results of the preliminary study of the Convolutional Neural Network (CNN) application on the ship structure, from 127 images obtained the highest number of labels is pitting corrosion, then general corrosion and the least is edge corrosion. The program design at the preliminary study stage is already able to detect corrosion with 3 categories but still has a low accuracy value. Where the test evaluation has an average accuracy of 0.3 and an average recall of 0.5. This is due to the low amount of data used as input for training and testing. Therefore, in the next stage, it is necessary to increase the number of data samples as input in the Convolutional Neural Network (CNN) process.