Real-Time Detection System for Defective Passion Fruit Using Visual Recognition with YOLOv5

Artificial intelligence is very important in production processes,as it allows data analysis and optimization of planning and control tasks to increase efficiency and reduce costs. To achieve this goal, the YOLOv5 model was used in a passion fruit juice production factory, standing out in the effectiveness of the detection and classification of fruit quality, even considering the inherent variability in passion fruit characteristics. The model was trained with approximately 500 images and achieved an accuracy of 92.869% and a sensitivity of 91.224%. Key challenges included the need to balance data from different categories of defects that could be present in the fruit and the high variability of its characteristics. To address these challenges, the study increased the number of images for specific categories to improve accuracy. Future work will focus on further optimizing the system by expanding the dataset size and refining the learning algorithms to keep the processing time within 30 minutes per session. This research provides a solid framework for the integration of advanced technological solutions in juice production, significantly improving the sustainability and efficiency of passion fruit processing. The results indicate a promising direction for the application of deep learning in this industry, offering scalable and effective methods to improve product consistency and reduce waste.