Research on a Product Quality Monitoring Method Based on Multi Scale PP-YOLO

Abstract

To monitor product quality in the production process in real time, this thesis proposes a quality monitoring model based on PaddlePaddle You Only Look Once (PP-YOLO). First, in the preprocessing stage, the data enhancement method and the K-means++ method are used to improve the robustness of the algorithm, and the generated anchor box can screen more refined features earlier. Second, ResNet50-vd with the deformable convolution idea is selected as the backbone of the detection model, the feature pyramid network structure and the composition of the loss function are improved, and the feature learning ability of the model is enhanced to enable it to detect multiple scales of defects. Finally, pruning is performed on the basis of the trained model to reduce the number of model parameters so that it can be deployed in industrial scenarios with limited hardware conditions. Experimental results show that the proposed quality monitoring model can meet the requirements for detection speed and accuracy in actual production, providing a new concept for the deployment of deep learning models in the industrial field.

Synthetic Generation and Detection of Crown-of-Thorns Starfish Instances retrival in Underwater Image

“…Variant PP-YOLOE-l achieves 51.4 mAP and 78.1 FPS, outperforming YOLOX in speed and accuracy [15].The thesis uses PP-YOLO to evaluate product quality in real time while also optimizing data processing and model efficiency. Achieves industrial precision and speed, demonstrating deep learning's potential for quality control [16].Enhanced PP-YOLO and improved Deep-SORT, with ResNet50 and margin loss, develop a real-time UAV tracking system (91.6%) that overcomes minor, featureless problems and interference issues [17].Traditional fire detection approaches have drawbacks; PP-YOLO, a cutting-edge object recognition model, uses computer vision and deep learning to enhance, promising early fire detection [18]. Study pioneers' real-time bivalve behavior monitoring for pollution utilizing four machine learning approaches, achieving anomaly identification without false alarms, demonstrating the possibility for automated aquatic pollution monitoring [19].…”

BoatNet: automated small boat composition detection using deep learning on satellite imagery

“…In this research, the YOLO framework was selected. It uses a multi-scale detection method, which enables it to detect objects at different scales and to adapt to changes in the size and shape of the objects being observed [ 88 ]. Besides, YOLO is highly effective in detecting small objects with high accuracy and precision [ 89 ].…”

Computer Vision-Based Fish Feed Detection and Quantification System

“…Betti and Tucci (2023), emphasized the importance of variation in the dataset to train the object detection model. According to Li et al (2021), there is a positive relationship between the number of datasets used and the accuracy level of the model in detecting objects. The larger the number of datasets used in training, the higher the accuracy level of the model in recognizing and detecting objects.…”

“…In this study, calculating the equations from the confusion matrix resulted in an accuracy of 0.82 or 82%, precision of 0.80, and recall of 0.84. According to Li et al (2021), an accuracy value of 1 indicates perfect classification, while a value of 0 indicates very poor classification. A precision value of 1 indicates that all predicted positive objects are correct, while a value of 0 indicates that no objects are correctly predicted as positive.…”