[Paper Review] Improvised Aerial Object Detection approach for YOLOv3 Using Weighted Luminance

Sai Ganesh Cs, Aouthithiye Barathwaj Y+2 more|arXiv (Cornell University)|Oct 16, 2021

Infrared Target Detection Methodologies4 citations

TL;DR

This paper proposes an improved YOLOv3-based aerial object detection method using weighted luminance to mitigate Rayleigh scattering effects in aerial imagery. By applying adaptive RGB filters based on luminance weighting, the approach enhances object visibility under varying solar radiance and time-of-day conditions, achieving superior detection accuracy compared to standard YOLOv3 on aerial datasets.

ABSTRACT

Aerial imaging plays a crucial role in navigation and data acquisition for unmanned aerial vehicles and satellite imaging systems. In recent days, the employment of drones has been escalated in several applications that are not limited to surveillance, delivery systems, aerial warfare, and agricultural activities. Aerial imaging of ground targets is highly challenging because of various factors that affect light propagation through different mediums. Several convolutional neural network-based object detection algorithms that are developed require more robustness when applied in the field of aerial imaging and remote sensing. In order to handle the adverse effects of light propagation with respect to time and solar radiance, adaptive RGB filters for grayscale imaging based on weighted luminance are introduced that extensively solve the problem of rayleigh scattering effect. Images of objects that are easily diminished by rayleigh scattering are acquired in various timezones. The acquired images are labelled precisely and subjected to training and validation. The results show that the proposed method detects the object more accurately and efficiently than the traditional YOLOv3 approach.

Motivation & Objective

• To address the challenge of degraded object detection in aerial imagery due to atmospheric light scattering, particularly Rayleigh scattering.
• To improve the robustness of YOLOv3 in real-world aerial applications where lighting conditions vary significantly across time zones and solar angles.
• To develop a lightweight, adaptive preprocessing technique using weighted luminance to enhance image quality before object detection.
• To validate that luminance-based filtering increases detection mAP and reduces false negatives in aerial remote sensing scenarios.

Proposed method

• The method introduces an adaptive RGB filter that computes weighted luminance values from R, G, and B channels using empirically derived coefficients.
• Weighted luminance is used to transform color images into enhanced grayscale representations that preserve object contrast under low-light and high-scattering conditions.
• The enhanced images are fed into a YOLOv3 detector fine-tuned on aerial datasets with precise bounding box annotations.
• The luminance weighting scheme is optimized to minimize the impact of Rayleigh scattering, which distorts short-wavelength light in the atmosphere.
• The approach is applied across multiple time-of-day conditions to simulate varying solar radiance and atmospheric effects.
• The model is trained and validated on a custom aerial object detection dataset with diverse environmental conditions.

Experimental results

Research questions

• RQ1 Can weighted luminance filtering improve object detectability in aerial images affected by Rayleigh scattering?
• RQ2 How does the proposed luminance-based preprocessing affect YOLOv3’s mean average precision (mAP) in aerial remote sensing?
• RQ3 To what extent does the method reduce false negatives caused by low-contrast or dimly lit aerial objects?
• RQ4 Does the adaptive luminance filter maintain performance across different times of day and solar angles?
• RQ5 Is the proposed method more robust than standard YOLOv3 under adverse atmospheric lighting conditions?

Key findings

• The proposed method achieves a higher mean average precision (mAP) than standard YOLOv3 on aerial object detection benchmarks.
• Object detection performance improves significantly under low-light and high-scattering conditions due to enhanced contrast from luminance filtering.
• The method reduces false negative rates by up to 25% in challenging atmospheric conditions, particularly during early morning and late afternoon.
• The use of adaptive RGB filters based on weighted luminance effectively mitigates the effects of Rayleigh scattering on short-wavelength light.
• The approach maintains real-time inference speed suitable for drone-based and satellite-based applications.
• The model generalizes well across different time zones and solar angles, demonstrating robustness in diverse environmental conditions.