An airline baggage feature perception network model was designed with PointNet++ as the benchmark and incorporating graph convolutional neural network and self-attention mechanism aiming at the problem that the configuration feature perception capability of airline baggage was inadequate under the demand of automatic baggage stacking handling. The local spatial attention module was introduced in the feature abstraction layer of the backbone network to extract associated spatial structure features of neighboring points in aviation baggage point cloud in order to perceive the intrinsic connection of its region feature space. Correlation between local features of airline baggage point cloud was learned through the global feature aggregation module to adaptively aggregate local features so as to form global contextual information. The recycling maxpooling layer was applied to recycle features from some discard points in the feature reduction process and collect baggage information at multiple levels, reducing information redundancy while retaining local and global feature activations with stark intensity. The experimental results showed that the average and overall accuracy of airline baggage classification were 94.68% and 96.32%, which were 6.53% and 5.07% improved over PointNet++, respectively. The airline baggage feature perception performance of the network model is better than other existing intelligent algorithms, which can provide accurate, reliable and effective input for airline baggage stacking space optimization and control.
Fig.1Experiment system of airline baggage feature collection
Fig.2Operation status of airline baggage feature collection system
Fig.3Classification of airline baggage
类别
表面材质
形状特点
方箱
纸板、塑料
规则长方体
硬箱
金属、塑料
光滑曲面、类长方体
软箱
织物、皮革
凹凸曲面、类长方体
硬包
金属、塑料
光滑曲面、类椭球体
软包
织物、皮革
凹凸曲面、类椭球体
异形
塑料保护套
球形、柱形不规则物体
Tab.1Basis for classification of airline baggage
Fig.4Structure of GACP network
Fig.5Structure of local spatial attention module
Fig.6Structure of global feature aggregation module
Fig.7Schematic diagram of recycling maxpooling
软硬件名称
软硬件配置
参数
数值
Operation system
Windows 10
Batch size
8
CPU
Intel Xeon E5-2680 v4
Input point
1 024
GPU
NVIDIA RTX 3080 Ti
Epoch
200
RAM
64 GB
Learning rate
0.001
Python+Pytorch
3.7. 0+1.7.1
Gamma
0.7
CUDA+CUDNN
11.0+8.1.0
Optimizer
Adam
Tab.2Experimental configuration of airline baggage feature perception
Fig.8Production method of extended dataset
算法
方箱
硬箱
软箱
硬包
软包
异形
mAcc/%
OA/%
t/s
PointNet
86.92
86.14
87.47
80.76
85.33
89.45
86.01
89.67
124.5
PointNet++
88.61
88.27
89.94
82.75
88.03
91.31
88.15
91.25
267.3
DGCNN
90.56
92.73
94.43
84.67
87.26
94.68
90.72
93.06
192.4
PointVGG
91.23
93.52
94.48
87.54
89.73
94.81
91.89
94.18
216.7
GACP（本文方法）
93.72
95.64
96.39
91.16
92.53
98.66
94.68
96.32
283.2
Tab.3Comparision of classification accuracy for different models
Fig.9Classification accuracy curves of different models
组数
LSA
GFA
RMP
mAcc/%
OA/%
A
×
×
×
88.15
91.25
B
√
×
×
90.07
92.61
C
√
√
×
93.45
95.73
D
√
√
√
94.68
96.32
Tab.4Ablation experiments of different modules
Fig.10Perceptual results of PointNet++
Fig.11Perceptual results of GACP
Fig.12Anomalies in luggage point cloud data
Fig.13Comparison of robustness experiment results
[1]
DAVID L. Baggage IT insights 2022 [R]. Brussels: SITA, 2022.
[2]
GAO Q, LIANG P Airline baggage appearance transportability detection based on a novel dataset and sequential hierarchical sampling CNN model[J]. IEEE Access, 2021, 9 (2): 41833- 41843
[3]
PENG C, GAO Q, LUO Q. Method on 3d reconstruction of airline baggage based on active laser projection of improved encoding [C]// IEEE CSAA Guidance, Navigation and Control Conference . Xiamen: IEEE, 2018: 1-7.
[4]
TANG C, ZHANG X, YU X, et al Target detection and classification based on LiDAR[J]. American Academic Scientific Research Journal for Engineering, Technology, and Sciences, 2018, 49 (1): 28- 39
[5]
MAZZEO P L, LIBETTA C, SPAGNOLO P, et al A siamese neural network for non-invasive baggage re-identification[J]. Journal of Imaging, 2020, 6 (11): 126
doi: 10.3390/jimaging6110126
[6]
CHEN R, ZHANG H, LI C, et al. LSDNN: local-salient deep neural network for baggage re-identification with material discerning [C]// Chinese Automation Congress . Shanghai: IEEE, 2020: 6344-6349.
[7]
ZHAO Q, MA H, LU R, et al. MVAD-Net: learning view-aware and domain-invariant representation for baggage re-identification [C]// Pattern Recognition and Computer Vision . Beijing: Springer, 2021: 142-153.
[8]
WU H, LUO Z, CAO D, et al. Attention and multi-grained feature learning for baggage re-identification[C]// Computer Supported Cooperative Work and Social Computing . Xiangtan: Springer, 2022: 460-472.
[9]
YANG H, CHU X, ZHANG L, et al QuadNet: quadruplet loss for multi-view learning in baggage re-identification[J]. Pattern Recognition, 2022, 126 (5): 108- 109
ZHANG Z, LI D, WU J, et al. MVB: a large-scale dataset for baggage re-identification and merged Siamese networks [C]// Pattern Recognition and Computer Vision . Xi’an: Springer, 2019: 84-96.
[12]
CAZAN C. Identification of surface material of baggage for self-service bag drop system [D]. Hong Kong: The University of Hong Kong, 2020.
[13]
QI C R, YI L, SU H, et al PointNet++: deep hierarchical feature learning on point sets in a metric space[J]. Advances in Neural Information Processing Systems, 2017, 30 (5): 78- 92
[14]
GAO Z, YAN J, ZHAI G, et al. Learning local neighboring structure for robust 3d shape representation [C]// Proceedings of the AAAI Conference on Artificial Intelligence , 2021, 35(2): 1397-1405.
[15]
VASWANI A, SHAZEER N, PARMAR N, et al Attention is all you need[J]. Advances in Neural Information Processing Systems, 2017, 30 (8): 5998- 6008
[16]
DU Z, YE H, CAO F A novel local-global graph convolutional method for point cloud semantic segmentation[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 15 (6): 1- 15
[17]
NIU Z, ZHONG G, YUE G, et al Recurrent attention unit: a new gated recurrent unit for long-term memory of important parts in sequential data[J]. Neurocomputing, 2022, 517 (4): 1- 9
[18]
CHEN J, KAKILLIOGLU B, REN H, et al. Why discard if you can recycle?: a recycling max pooling module for 3d point cloud analysis [C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition . New Orleans: IEEE, 2022: 559-567.
[19]
LI R, LI X, HENG P A, et al. PointAugment: an auto-augmentation framework for point cloud classification [J]. IEEE/CVF Conference on Computer Vision and Pattern Recognition . Seattle: IEEE, 2020: 6378-6387.
[20]
CHARLES R Q, HAO S, MO K C, et al. PointNet: deep learning on point sets for 3d classification and segmentation [C]// IEEE Conference on Computer Vision and Pattern Recognition . Honolulu: IEEE, 2017: 77-85.
[21]
WANG Y, SUN Y, LIU Z, et al Dynamic graph CNN for learning on point clouds[J]. ACM Transactions on Graphics, 2019, 38 (5): 146- 152
