Abstract
In automatic loading and unloading systems, it is crucial to accurately detect the locations of boxes inside trucks in real time. However, the existing methods for box detection have multiple shortcomings, and can hardly meet the strict requirements of actual production. When the truck environment is complex, the currently common models based on convolutional neural networks show certain limitations in the practical application of box detection. For example, these models fail to effectively handle the size inconsistency and occlusion of boxes, resulting in a decrease in detection accuracy. These problems seriously restrict the performance and reliability of automatic loading and unloading systems, making it impossible to achieve ideal detection accuracy, speed, and adaptability. Therefore, there is an urgent need for a new and more effective box detection method. To this end, this paper proposes a new model, HYFF-CB, which incorporates key technologies such as a location attention mechanism, a fusion-enhanced pyramid structure, and a synergistic weighted loss system. After real-time images of a truck were obtained by an industrial camera, the HYFF-CB model was used to detect the boxes in the truck, having the capability to accurately detect the stacking locations and quantity of the boxes. After rigorous testing, the HYFF-CB model was compared with other existing models. The results show that the HYFF-CB model has apparent advantages in detection rate. With its detection performance and effect fully meeting the actual application requirements of automatic loading and unloading systems, the HYFF-CB model can excellently adapt to various complex and changing scenarios for the application of automatic loading and unloading.