Artificial intelligence (AI) has become the new technique for solving the most complicated challenges related to big data, image recognition, object detection, and other classification and prediction issues due to the fourth industrial revolution.

This article introduces the development and implementation of a zinc-plated component recognition system within a manufacturing process using deep learning (DL) techniques.This paper aims to train and evaluate different DL algorithms to detect five zinc-plated components and one assembly tray under different ambient light conditions and finishings. The proposed method begins with creating a custom dataset with six other classes that match the assembly components. Then, the image augmentation technique is applied to a custom dataset obtaining 5.5K images. Six deep neural networks have been benchmarked: Data-efficient image Transformer (DeiT), ResNet101V2, Xception, InceptionV3,ResNet152V2, and EfficientNetV2S. The training stage is performed with the k-folded method and K=6, and a progressive data size between different runs with an incremental of 10%. In thisstudy, DeiT was the model with the highest F1-Score, rated at 99.89%, while EfficientNetV2Sachieved the lowest F1-Score: 77.18%. Therefore, the first four algorithms can be adopted as good classifier models for zinc-plated component detection in the industrial manufacturing production.