Design And Development Of A Supervised Learning Model For Detection Of Colours In Objects Using Convolution Neural Network

Abstract

Corrosion is one big problem affecting the several networks in building. Due to corrosion approximately 20% of the sewer network in the world is damaged due to corrosion. The only way to assess the damage is through visual inspection. There are many ways which can be used to inspect the corrosion caused in the sewer pipe. One of the advanced techniques used can be by using Artificial Intelligence (AI). AI can assist in inspecting the corroded pipes and also provide better objective classification. This technique reduces cost and can be directly integrated with digital asset management systems. This paper aims at automated detection and localization of key symptoms such as molds, deterioration and stains by the application of convolution neural network (CNN). A preprocessing model will be developed to classify the images depending on whether it contains rust or not. The second process will help in localizing the rusted region. The models will be trained with a large set of training sets and will be evaluated on different parameters. Preprocessing is performed by a supervised learning which helps in classification of the rust. The data will be classified using binary classification, in this classification model we will be using ROC (Receiver Operating Characteristics) AUC (Area under the Curve) to check the performance of a binary or multi class classification. In addition to this Tensor flow object detection API will also be used to detect and classify the images.