[Paper Review] Grayscale Based Algorithm for Remote Sensing with Deep Learning.

Sai Ganesh Cs, Aouthithiye Barathwaj S. R. Y+1 more|arXiv (Cornell University)|Oct 16, 2021

Advanced Image Fusion Techniques21 references4 citations

TL;DR

This paper proposes a grayscale-based YOLOv3 algorithm to improve remote sensing target detection by mitigating atmospheric light distortion effects such as Rayleigh scattering. By converting RGB images to grayscale before inference, the method enhances detection accuracy and robustness compared to standard YOLOv3, particularly under challenging atmospheric conditions.

ABSTRACT

Remote sensing is the image acquisition of a target without having physical contact with it. Nowadays remote sensing data is widely preferred due to its reduced image acquisition period. The remote sensing of ground targets is more challenging because of the various factors that affect the propagation of light through different mediums from a satellite acquisition. Several Convolutional Neural Network-based algorithms are being implemented in the field of remote sensing. Supervised learning is a machine learning technique where the data is labelled according to their classes prior to the training. In order to detect and classify the targets more accurately, YOLOv3, an algorithm based on bounding and anchor boxes is adopted. In order to handle the various effects of light travelling through the atmosphere, Grayscale based YOLOv3 configuration is introduced. For better prediction and for solving the Rayleigh scattering effect, RGB based grayscale algorithms are proposed. The acquired images are analysed and trained with the grayscale based YOLO3 algorithm for target detection. The results show that the grayscale-based method can sense the target more accurately and effectively than the traditional YOLOv3 approach.

Motivation & Objective

• Address the challenge of atmospheric light distortion in satellite-based remote sensing, particularly Rayleigh scattering effects.
• Improve target detection accuracy in remote sensing by reducing noise and complexity from RGB color channels.
• Enhance the performance of YOLOv3 in remote sensing applications through grayscale preprocessing.
• Evaluate the effectiveness of grayscale transformation in improving model generalization and detection robustness under variable atmospheric conditions.

Proposed method

• Adopt YOLOv3 as the base object detection framework due to its efficiency and anchor box-based bounding box prediction.
• Convert input RGB remote sensing images to grayscale prior to model inference to reduce spectral complexity and noise.
• Apply grayscale transformation to mitigate the impact of atmospheric scattering, especially Rayleigh scattering, on image quality.
• Train the grayscale-based YOLOv3 model using supervised learning with pre-labeled target data for classification and localization.
• Utilize anchor boxes and feature pyramid networks in YOLOv3 to detect objects at multiple scales in remote sensing imagery.
• Process satellite images through the grayscale YOLOv3 pipeline to predict bounding boxes and class labels for ground targets.

Experimental results

Research questions

• RQ1 Can grayscale preprocessing improve the accuracy of object detection in remote sensing imagery affected by atmospheric light distortion?
• RQ2 How does the grayscale-based YOLOv3 model compare to standard YOLOv3 in detecting ground targets under varying atmospheric conditions?
• RQ3 To what extent does grayscale transformation reduce the impact of Rayleigh scattering on remote sensing image quality?
• RQ4 Does the use of grayscale images enhance model robustness and generalization in remote sensing applications?

Key findings

• The grayscale-based YOLOv3 method achieves higher detection accuracy than standard YOLOv3 in remote sensing scenarios.
• Grayscale preprocessing effectively reduces the influence of atmospheric light scattering, particularly Rayleigh scattering, on image clarity.
• The model demonstrates improved robustness in detecting ground targets under varying atmospheric conditions due to reduced spectral noise.
• The use of grayscale images enhances the model’s ability to focus on structural and textural features critical for target detection.
• The results confirm that grayscale transformation leads to more effective feature learning in remote sensing applications.
• The proposed method outperforms traditional YOLOv3 in target detection tasks involving satellite imagery with atmospheric interference.