"Application of Computer Vision and Artificial Intelligence for Briquette Detection and Classification"

"The dataset consisted of 51 labeled images of briquettes captured in a laboratory environment using three different backgrounds: white, black, and a treadmill surface. These images were divided into three subsets: approximately 63% (32 images) for training, 19% (10 images) for validation, and 19% (9 images) for testing. The dataset included two object classes: briquete ruim (defective briquettes) and briquette bom (in tact briquettes), as defined in the configuration file data.yaml. Validation was performed continuously during training, allowing the model to be iteratively evaluated and improved. The main performance metric used was the mean Average Precision (mAP@0.5), which reached a peak of approxi mately 0.84593, indicating strong classification capability. False positive and false negative rates were also monitored to ensure the model could distinguish accurately between intact and defective briquettes. Inference speed was evaluated throughout the process and confirmed to be compatible with real-time industrial requirements."