Deep Learning Approach for Object Classification on Raw and Reconstructed GBSAR Data

Abstract

The availability of low-cost microwave components today enables the development of various high-frequency sensors and radars, including Ground-based Synthetic Aperture Radar (GBSAR) systems. Similar to optical images, radar images generated by applying a reconstruction algorithm on raw GBSAR data can also be used in object classification. The reconstruction algorithm provides an interpretable representation of the observed scene, but may also negatively influence the integrity of obtained raw data due to applied approximations. In order to quantify this effect, we compare the results of a conventional computer vision architecture, ResNet18, trained on reconstructed images versus one trained on raw data. In this process, we focus on the task of multi-label classification and describe the crucial architectural modifications that are necessary to process raw data successfully. The experiments are performed on a novel multi-object dataset RealSAR obtained using a newly developed 24 GHz (GBSAR) system where the radar images in the dataset are reconstructed using the Omega-k algorithm applied to raw data. Experimental results show that the model trained on raw data consistently outperforms the image-based model. We provide a thorough analysis of both approaches across hyperparameters related to model pretraining and the size of the training dataset. This, in conclusion, shows how processing raw data provides overall better classification accuracy, it is inherently faster since there is no need for image reconstruction and it is therefore useful tool in industrial GBSAR applications where processing speed is critical.

Ground based SAR system for object classification with parameter optimization based on deep learning feedback algorithm

“…The averaged signal from each step (with the size of 1024 points) is stored to Raspberry Pi which means that the result matrix of one GBSAR measurement has dimensions {number of steps}x1024. GBSAR-Pi scheme with pin connections is given in Figure 1, while the whole system is described in more details in 7 .…”

“…Hence, the classification is based on raw GBSAR data, rather than on reconstructed images. 7 As mentioned, in each of the sensor's positions signal with 1024 frequency points is stored. Hence, one spatial dimension of the matrices is always 1024.…”

“…The classification model is applied on raw GBSAR data matrices 7 (instead of being applied on reconstructed radar images) and is based on ResNet18 Convolutional Neural Network architecture. The well-known advantage of the architecture lies in its simplicity and efficiency in classification trained on data in form of matrix which is significant for implementation on low-power devices such as Raspberry Pi.…”

“…The combinations of such algorithms with Ground Based Synthetic Aperture Radar (GBSAR) include deformation forecasting, [1][2][3] terrain mapping 4,5 and object classification. 6,7 Most of them use data obtained from radar as direct input to ML model implemented either on the microcomputer connected with the radar or on the remote computer. However, when implemented together, radar and ML model can share information in order to improve ML accuracy results or even energy efficiency of the system.…”

Near-distance raw and reconstructed ground based SAR data

“…Such format can also be utilized in raw-data based object classification, while reconstructed images in image-data based object classification. In [1] the comparison between those two approaches based on ResNet18 architecture is given. However, various deep learning models based on both raw and reconstructed radar data can be implemented using this dataset.…”