Defect Detection Of Automotive Composite Leather Based On Nano Det-Plus

In recent years,the increase in living standards has led to a continuous growth in the number of privately owned cars.Consumers not only pay attention to the performance and quality of automobiles but also increasingly value the quality of automotive interior decorations.As a result,the quality of automotive interior decorations has become one of the important factors for consumers to consider,and the demand for automobiles has gradually shifted from "quantitative growth" to "qualitative growth".Leather,with its excellent characteristics such as breathability,ease of maintenance,and inherent elasticity,has become one of the main materials for automotive interiors,and its quality requirements have become increasingly stringent.However,various defects may occur during the production process of automotive composite leather,such as defects in raw materials or human error or processing technology issues.Therefore,before cutting and processing,defects on automotive composite leather need to be detected,located,and treated to ensure the quality of automotive composite leather products.This paper applies computer vision and deep learning technologies and conducts research on a defect detection method for automotive composite leather based on the Nano Det-plus object detection model.The main work of this study includes the following aspects:1.By consulting relevant literature and conducting practical investigations,the current development status of traditional leather defect detection algorithms and deep learning-based object detection algorithms,along with their advantages and limitations,both domestically and internationally,were analyzed.The demands in the field of automotive composite leather production were also examined.Consequently,the single-stage anchor-free object detection algorithm Nano Det-plus was introduced into the task of defect detection for automotive composite leather,based on academic perspectives.2.Through on-site investigations and examinations,samples of defects in automotive composite leather were collected.A platform for defect image collection was established to capture and perform basic preprocessing on the samples.Considering the insufficient data volume that can significantly affect the effectiveness of object detection algorithms,this study employed data augmentation methods to expand and annotate the collected data of defects in automotive composite leather,thus constructing a dataset for automotive composite leather defect detection.3.This study elaborates on the network structure of the Nano Det-plus object detection model.To address the issue of high similarity between defects and the background in automotive composite leather,several lightweight neural networks with superior performance were employed in the feature extraction module to extract features from the dataset of defects in automotive composite leather.A comparison of the detection results from different models was conducted to select the most suitable one.Performance comparisons were also conducted on the constructed dataset of defects in automotive composite leather with excellent object detection models such as the YOLO series.4.This study addresses the increasing number of defect samples generated during the actual production process,including the potential involvement of new categories.It applies fine-tuning and knowledge distillation techniques to incrementally improve the Nano Detplus model.Experimental evidence is provided to demonstrate the effectiveness of incremental learning.5.To reduce costs in practical applications,the trained model in this study was deployed on an embedded development board(NVIDIA Jetson TX2)with lower costs and fewer resource requirements.An appropriate defect selection strategy was chosen,and a user interface program was developed using Py Qt5 to enable control and data management for real-time detection of defects in automotive composites.