Target Recognition in Infrared Circumferential Scanning System via Deep Convolutional Neural Networks

Area of Science:

• Computer Vision
• Artificial Intelligence
• Defense Technology

Background:

• Long-time surveillance over large fields of view is essential for environmental awareness, particularly in defense.
• Automated target recognition, comprising detection and identification, is critical for modern defense systems.
• Existing deep convolutional neural network (DCNN) methods for target recognition require extensive annotated datasets, which are scarce for infrared imagery.

Purpose of the Study:

• To develop an end-to-end deep convolutional neural network (DCNN)-based method for target recognition in infrared circumferential scanning systems (IRCSS).
• To address the challenge of limited annotated infrared datasets by creating a new dataset and employing data augmentation and transfer learning.
• To improve the localization performance of target recognition using a smoother L1 loss function.

Main Methods:

• Development of a DCNN-based approach for end-to-end target recognition within IRCSS.
• Creation of a novel infrared target recognition dataset to mitigate data scarcity and improve scene adaptability.
• Implementation of data augmentation, cross-domain transfer learning, and a smoother L1 loss function for bounding box regression.

Main Results:

• The proposed DCNN method achieved an 82.7 mAP (mean Average Precision).
• The system demonstrated effective end-to-end infrared target recognition.
• High accuracy in both target detection and identification was accomplished.

Conclusions:

• The developed DCNN method provides a highly effective solution for end-to-end infrared target recognition in IRCSS.
• The creation of a dedicated dataset and the application of advanced training strategies enhance algorithm performance and adaptability.
• The study significantly advances automated target recognition capabilities for defense applications.