Object classification and recognition method in railway material storage area based on channel weight allocation
There are a wide variety of goods in large quantities in railway material warehouses,making manual han-dling difficult.It is urgent to construct unmanned warehouses to elevate the level of material management.However,railway material warehouses often involve emergency repair tasks,high goods dynamics,humans and machines mixed together,and workers,goods,and vehicle bodies blocking each other,significantly affecting object recognition.Therefore,how to accurately identify common objects in the warehouse and make timely task adjustments has become an urgent issue to be improved in the intelligent operation of railway warehouses.Based on the YOLOv8 model,SE(Squeeze and Excitation)attention module was added to the backbone network to optimize the loss function,and a YOLOv8-S railway object recognition model used in maintenance and storage areas was proposed.By introducing the SE module,the YOLOv8 network can achieve primary and secondary priority by assigning weights between different channels,improve the model's attention to features and further enhance its classification ability.The optimized loss function calculated IoU by minimizing the point distance between the predicted bounding box and the truthy bounding box,simplifying the calculation process.The images of four types of objects,namely forklifts,goods,staff,and shelves,that appear frequently in railway maintenance warehouses were collected as a dataset.To ensure the quality of the dataset,MSE mean square error was used to remove redundant images,with a total of 1000 images obtained.The four types of objects were labeled and fed into a training model.The test results show that the model has a recall rate of 83.7％and an average mean accuracy of 87.6％.Compared with the original YOLOv8,YOLOv8-S has in-creased its recall rate by 3.8％and average mean accuracy by 4.3％.The results indicate that the YOLOv8-S net-work can accurately recognize common objects in railway maintenance warehouses.
object recognitionrailway maintenance and storage areaYOLOv8 algorithmattention mechanism
NETL
NSTL
万方数据
维普
