Application of Single Shot MultiBox Detector in Logistics Safety Testing

2019 2nd International Conference on Advanced Materials, Intelligent Manufacturing and Automation

Xia, Xiaoling^1 ; Shi, Xin^1 ; Lu, Qinyang^1 ; Fan, Changqi^2

College of Computer Science, Donghua University, Shanghai

201620, China^1

I-Soft Co., Ltd, China^2

关键词: Application method;    Convolution neural network;    Cubic-spline interpolation;    High-accuracy;    Internet economy;    Logistics industry;    Logistics package;    Negative samples;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/569/2/022002/pdf
DOI  :  10.1088/1757-899X/569/2/022002

来源: IOP

【 摘 要 】

With the rapid development of the Internet economy, the load of the logistics industry has gradually increased, and people are paying more and more attention to the detection of restricted items in logistics packages. Using Convolution Neural Network can quickly and accurately identify the restrictions in the package, thus minimizing the pressure on personnel in logistics monitoring. This paper introduces the application method of SSD-based target detection model in the detection of logistics goods restriction products. Firstly, it analyses the characteristics of data sets and preprocesses the data and expounds the feasibility of SSD model in this respect. Different from the traditional upsampling method, this paper uses cubic spline interpolation to adjust the image resolution and expands the number of positive samples to balance the number of positive and negative samples. Next, batch normalization is added after the convolution layer and Softmax is replaced with multiple logistic classifiers so that the performance of the model can be improved. Finally, by comparing the results of SSD and Fast R-CNN, this paper proves that SSD, which obtains a mAp of 0.476 at more than 30 FPS, is better on this dataset. It owns both high accuracy and real-time speed.