Convolutional Neural Network Models for Scattering Pattern Recognition of Scanning Electron Microscopy Images

Morphology is an important research method that reveals the properties of materials. Each material has a unique scattering pattern that it depends on the kind of material, preparation method, and the composition of the blends. These scattering patterns are captured to images by the scanning electron microscopy (SEM). Since there are a lot of scattering patterns, it is very hard to interpret and recognize the scattering pattern of any materials. This work applied the CNN model to recognize the scattering pattern in the SEM images by creating the appropriate model that suits for recognize these images. The SEM image dataset was built and characterized the model structure. The experiment results reveal that the CNN model could significantly recognized the SEM images. In which way, this work could solve the problem of false interpretation of the material morphology and improve the efficiency of classification obviously. Moreover, the CNN model could recognize the scattering pattern with an accuracy at 97.02% and 2.98% of error. A contribution of this study is a new utilization of SEM images classification and recognition, which utilize the laboratory experiments. This work also created the SEM images dataset which is a new knowledge for enhancing the morphology study in the material engineering.