| In recent years,the development of the Chinese express industry has made remarkable achievements,and China has the largest express service market in the world.However,the problem of rough handling of express parcels has not been resolved properly.Rough handling of express parcels greatly increases the risk of shipment damage,affects the image of the industry negatively,and causes serious social problems such as over-packaging.Realizing the accurate and efficient recognition of rough handling of express parcels is the major prerequisite for solving this problem.On this basis,targeted measures such as adding auxiliary equipment and strengthening personnel training can be adopted to avoid the occurrence of rough handling of express parcels.Therefore,this article has carried out the following research works:Through the study of relevant regulations,standards and references,and the field investigation of express companies,three typical classes of rough handling of express parcels are summarized,namely,dropping,throwing,and kicking.For these three typical classes of abnormal logistics operations,this article discusses the value and significance of recognizing them.A detection and recognition method for rough handling of express parcels based on the acceleration sensor and deep learning is proposed.Moreover,a detection and recognition system for rough handling of express parcels is designed and implemented based on this method.In this system,the data collection terminal collects the three-axis acceleration data of the express parcel in real-time and then uses the GPRS module to upload the potentially abnormal data to the server for feature extraction and pattern recognition.Finally,the time,location,type,and the other information of the rough handling of express parcels are matched and saved in the database for users to query.At the hardware level,a data collection terminal for the three-axis acceleration status of the express parcel is designed and implemented.At the algorithm level,a complete set of intelligent recognition algorithms for rough handling of express parcels is designed and implemented.At first,the algorithms intercept the potentially abnormal sample,add windows to the intercepted three-axis acceleration data,and extract the seven features of the data in the windows including mean,variance,kurtosis,skewness,dynamic range,short-term energy,and zero-crossing rate.After these processings,normalization is performed,and the number of time windows is aligned,so as to obtain the traditional feature matrix shaped like 3 axes×50 time windows×7 features.Then the normalized traditional feature matrix is fed into a CNN-GRU(Convolutional Neural Networks-Gated Recurrent Unit)fusion model with the CDCE(Channel Dense--Concatenation-Excitation)channel attention blocks.And the channel,space,and time abstract relationship features of the matrix are extracted.Finally,the recognition result is obtained.Based on the traditional SE(Squeeze-Excitation)channel attention block,combined with the characteristics of the input data of the deep learning model under the application background of this article,an improved channel attention block(CDCE)is designed.It replaces the global pooling operation in the SE block with the sub-channel full connections and adjusts the subsequent layers accordingly to achieve finer channel weight calculations without significantly increasing the complexity.A three-axis acceleration state data set of express parcels under logistics operations was collected,sorted,and published.Using this data set can not only train and test the recognition model proposed in this article but also has values for extended research in this field.Several actual delivery tests further prove the feasibility and practicability of the method proposed in this paper.And the statistical analysis of the detection and recognition results of actual delivery tests reveals that the universality of rough handling of express parcels in the industry at present. |
PDF Full Text Request