CLASSIFICATION OF CONSTRUCTION ORDERING DOCUMENTS BY CNN AND APPLYING ROAD MAP INFORMATION GATHERING

Abstract

 If the information generated by routine work can be used for extraction of construction work whose conditions are designated, and the extraction can be done automatically, the efficiency will be significantly improved compared to creating a database and manual work. In this report, we picked up construction ordering documents as information to be used and showed the results and evaluation of experiments in which each page type is automatically classified into 8 classes by CNN (convolutional neural network). Through experiments, it was possible to extract a core page in tender notice or a covera page in implementation design documents from Shizuoka Prefecture's general civil engineering ordering documents and obtain an outline of the construction with 50 training data per class. The classification performance is improved by increasing the number of data, and it can be applied to the classification of construction ordering documents for road map updates. This method is expected to be applied to construction deliverables such as electronic ones.