This article is based on the motivation of turbine migration, selects high-dimensional data from the data set of WT (wind turbine) blades, overfitting and strict feature selection are the three keys to solving the problem of NN (neural network) computing systems. For the above-mentioned conflicting problem, in this study, the characteristics of the WT including the similarity in the color of WT blades and the difference in shape are deployed to evaluate the performance of the machine learning system. Therefore, the pre-processing with Bilateral filter is applied to join with the SS (selective synthesizer) of blade fouling patterns are proposed where adopts the framework of ResNet50 to examine the computational efficiency in the study. The experimental results verify that according to the feature selection after the introduction of the SS method, the accuracy rate of the NN model can reach higher than 92%. Finally, it is found that when the proposed CC (correlation coefficient) and SS are combined, the image pre-processing of the machine learning image data model can show the most significant feature selection performance. For the purpose of data validation, the YOLO (Only Look Once) deep learning framework. The reason for using YOLOv4-Tiny is that the current YOLO framework can obtain a compromise balance and practical recognition in terms of affecting the recognition speed and accuracy rate. Moreover, it has been integrated with the edge computing hardware Nvidia Jetson Nano.