Identify Target Area of Panel for Spraying Using Convolutional Neural Network

Implementation of Artificial Intelligence in the paint spray process in the automobile industry can increase the efficiency of the paint spray process and reduce waste disposal. This project presents a semantic segmentation with a trained Convolutional Neural Network (CNN) implemented in the paint spray process as it can be used to predict and identify the target area that needs to be sprayed. A dataset contains different types of cars annotated with the car parts. A series CNN trained to classify the types of cars. There are 16 different semantic CNNs with different architecture and data stores trained to compare the result. The last step is to develop a system to identify the spray area. The result showed the accuracy of series CNN is 1, and the best way to train semantic CNN, which trained according to each type of car with the architecture of ResNet-50 with a data store consists of a more class. The validation result of ResNet-50_Type_01, ResNet-50_Type_02 and ResNet-50_Type_03 are 93.8162%, 90.5214% and 91.8023% which all have exceeded 85% and the Mean IoU of ResNet-50_Type_01, ResNet-50_Type_02 and ResNet-50_Type_03 are 0.8456, 0.8392 and 0.8263.