摘要
The inventory counting of silver ingots plays a key role in silver futures.However,the manual inventory counting is time-consuming and labor-intensive.Furthermore,the silver ingots are stored in warehouses with harsh lighting conditions,which makes the automatic inventory counting difficult.To meet the challenge,we propose an automatic inventory counting method integrating object detection and text recognition under harsh lighting conditions.With the help of our own dataset,the barcode on each silver ingot is detected and cropped by the feature pyramid network(FPN).The cropped image is normalized and corrected for text recognition.We use the PSENet+CRNN(Progressive Scale Expansion Network,Convolutional Recurrent Neural Network)for text detection and recognition to obtain the serial number of the silver ingot image.Experimental results show that the proposed automatic inventory counting method achieves good results since the accuracy of the proposed object detection and text recognition under harsh lighting conditions is near 99%.
基金
Supported by the National Key R&D Program of China(2019YFE0190500)
the State Key Program of National Nature Science Foundation of China(61936001)
the Natural Science Foundation of Shanghai(20ZR1420400)。