Remote Sensing Imagery Object Detection Model Compression via Tucker Decomposition
Lang Huyan,
Ying Li,
Dongmei Jiang,
Yanning Zhang,
Quan Zhou,
Bo Li,
Jiayuan Wei,
Juanni Liu,
Yi Zhang,
Peng Wang,
Hai Fang
Affiliations
Lang Huyan
School of Computer Science, National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology, Shaanxi Provincial Key Laboratory of Speech & Image Information Processing, Northwestern Polytechnical University, Xi’an 710129, China
Ying Li
School of Computer Science, National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology, Shaanxi Provincial Key Laboratory of Speech & Image Information Processing, Northwestern Polytechnical University, Xi’an 710129, China
Dongmei Jiang
School of Computer Science, National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology, Shaanxi Provincial Key Laboratory of Speech & Image Information Processing, Northwestern Polytechnical University, Xi’an 710129, China
Yanning Zhang
School of Computer Science, National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology, Shaanxi Provincial Key Laboratory of Speech & Image Information Processing, Northwestern Polytechnical University, Xi’an 710129, China
Quan Zhou
School of Computer Science, National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology, Shaanxi Provincial Key Laboratory of Speech & Image Information Processing, Northwestern Polytechnical University, Xi’an 710129, China
Bo Li
Key Laboratory of Science and Technology on Space Microwave, CAST Xi’an, Xi’an 710100, China
Jiayuan Wei
Key Laboratory of Science and Technology on Space Microwave, CAST Xi’an, Xi’an 710100, China
Juanni Liu
Key Laboratory of Science and Technology on Space Microwave, CAST Xi’an, Xi’an 710100, China
Yi Zhang
Key Laboratory of Science and Technology on Space Microwave, CAST Xi’an, Xi’an 710100, China
Peng Wang
Key Laboratory of Science and Technology on Space Microwave, CAST Xi’an, Xi’an 710100, China
Hai Fang
Key Laboratory of Science and Technology on Space Microwave, CAST Xi’an, Xi’an 710100, China
Although convolutional neural networks (CNNs) have made significant progress, their deployment onboard is still challenging because of their complexity and high processing cost. Tensors provide a natural and compact representation of CNN weights via suitable low-rank approximations. A novel decomposed module called DecomResnet based on Tucker decomposition was proposed to deploy a CNN object detection model on a satellite. We proposed a remote sensing image object detection model compression framework based on low-rank decomposition which consisted of four steps, namely (1) model initialization, (2) initial training, (3) decomposition of the trained model and reconstruction of the decomposed model, and (4) fine-tuning. To validate the performance of the decomposed model in our real mission, we constructed a dataset containing only two classes of objects based on the DOTA and HRSC2016. The proposed method was comprehensively evaluated on the NWPU VHR-10 dataset and the CAST-RS2 dataset created in this work. The experimental results demonstrated that the proposed method, which was based on Resnet-50, could achieve up to 4.44 times the compression ratio and 5.71 times the speedup ratio with merely a 1.9% decrease in the mAP (mean average precision) of the CAST-RS2 dataset and a 5.3% decrease the mAP of the NWPU VHR-10 dataset.
