Remote Sensing Imagery Object Detection Model Compression via Tucker Decomposition

Abstract

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.

How Satellite Imaging and Deep Learning are Influenced by Tensor Decompositions: A Review

“…This method has been effectively used with several DL architectures, including convolutional networks and restricted Boltzmann machines, either by compressing the entire design or by applying TDs to specific layers. The integration of TDs with DL models, specifically focusing on modifying input aggregation functions using tensors within artificial neurons would explain how this approach can capture higher-order relationships in structured data, such as tree structures, and highlights using TDs as a tradeoff between simplicity and complexity in neural aggregation [27].…”

“…Notably, the use of TDs in DL models is an emerging research area that shows promise [27]. While tensor factorization is a well-established method for multi-way data analysis, tensor decompositions are also being explored for the enhancement of the expressiveness of neural representations.…”

“…The research themes present intriguing challenges that can enhance the performance of DL models and provide deeper insights into their functioning [27]. The question of whether tensorization should only have an impact on the forward phase of neural models (i.e., computing neural activity) or if it also has significant implications for the backward phase (i.e., learning) is one such difficult open problem brought about using tensorized neural layers.…”

“…TDs are fundamental components in modern DL architectures, serving as the building blocks for various operations like convolution and attention [27]. By incorporating tensor decompositions, deep neural networks can reduce the number of parameters, making them more efficient, robust, and capable of preserving the underlying structure in the data.…”

Compressive Sensing in Image/Video Compression: Sampling, Coding, Reconstruction, and Codec Optimization

“…However, the impact on performance under the scenario of a microcontroller operating at a higher frequency, and consequently exhibiting increased consumption, remains an area that requires further investigation in the future. Last but not least, CS, with its capability to substantially reduce the data representation size while preserving essential information, holds promising prospects for applications in many high-level vision tasks, such as object detection [109,110], semantic segmentation [111], and image classification [112,113]. It is imperative to underscore that the potential of CS in these domains is far from fully realized, and there exists substantial room for further advancements and breakthroughs.…”