Seeing Clearly: How Tech Boosts Tiny Object Detection in Blurry Infrared Pics

Infrared (IR) imaging is a big deal. It helps in many areas like spotting objects, checking factories, and even medical scans. But there's a catch: IR images are often blurry and noisy. This makes it tough for regular object detection models to work well.

A New Approach

To tackle this, a new method has been developed. It combines two key technologies:

Super-resolution
Improved YOLOv8 model

LightweightSRNet

The first part, called LightweightSRNet, makes blurry IR images clearer without using too much computer power.

HG-MHA

The second part, called HG-MHA, helps the model focus on the important stuff and ignore the noise.

Advanced Features

SC-BiFPN Module: Mixes different levels of details to make small objects easier to spot.
C2f-Ghost-Sobel Module: Helps detect edges and details quickly, making the whole process faster.

Test Results

Tests on the HIT-UAV dataset showed great results:

Recall rate went up from 70.23% to 80.51%
Mean average precision (mAP) improved from 77.48% to 83.32%

This means the model is much better at finding small objects in IR images.

Real-World Impact

This new method could be a game-changer for real-world applications. It makes IR imaging more reliable and efficient. The best part? The code and datasets used in this study are available for anyone to check out and use.