METHOD FOR IDENTIFYING AUTOMATIC REGION OF NON-METALLIC INCLUSION AND INCLUSION DETERMINATION SYSTEM USING THE SAME

To provide a technique of automatically and accurately performing a divisional identification of an image region of an inclusion of a type to be an evaluation target by deep learning in the determination of the size data of the largest inclusion, from an optical microscopic image, necessary to predict the size of the inclusion by an extreme value statistics method.SOLUTION: In the method of automatically and accurately performing a divisional identification of an image of a non-metallic inclusion into regions according to the types of inclusions, the feature amount of an input image is extracted when an optical microscopic image of a steel material of which region of inclusions is not identified (unknown) is referred to by using a learning model which an electronic calculator generated by a machine learning, using teacher data based on a group of learning datasets having plural pair of data, each including an optical microscopic image, as raw data, of a steel material including non-metallic inclusions to be a determination target and a color image, as correct data, of the inclusions to be a determination target are painted differently according to the types of inclusions, the inclusion types are determined, and the inclusions are painted according to the type of the inclusions.SELECTED DRAWING: Figure 1 【課題】極値統計法による介在物の大きさ予測において必要となる最大介在物の大きさデータをその光学顕微鏡画像から求めるにあたり、評価対象とする種類の介在物の画像領域をディープラーニングにより自動で精度良く分割識別する技術を提供する。【解決手段】生データである判別対象の非金属介在物が含まれる鋼材の光学顕微鏡画像と、この判別対象となる介在物が各種介在物系毎に塗り分けされた正解データである着色画像とのデータ対を複数対備えた一群の学習データセットに基づいた教師データを用いて電子計算機に機械学習で生成させた学習モデルを用いて、介在物の領域が未識別（未知）の鋼材の光学顕微鏡画像が照会されたときに入力画像の特徴量を抽出し、介在物系を判別し、介在物系毎に塗り分ける、非金属介在物の画像について各種介在物系毎の領域に自動で分割識別する介在物判別方法。【選択図】 図１