Target classification of multislit streak tube imaging lidar based on deep learning

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Multislit streak tube imaging lidar (MS-STIL) systems face challenges in object identification due to image reconstruction impacts.
• Improving the efficiency of MS-STIL systems is crucial for accurate target recognition.

Purpose of the Study:

• To develop and evaluate an object classification method utilizing MS-STIL echo data.
• To enhance the identification efficiency of MS-STIL systems by bypassing complex image reconstruction.

Main Methods:

• A streak image dataset of 240 outdoor targets across 6 categories was created.
• A ResNet-based deep learning model was employed for streak image classification.
• Classification performance was compared between streak images and reconstructed depth images.

Main Results:

• The proposed method achieved high classification accuracy, reaching 99.42% for 6 classes.
• Accuracy decreased to 67.64% for 20 classes, indicating a challenge with increased complexity.
• Expanding the dataset improved accuracy to 85.35% for 20 classes, demonstrating scalability.

Conclusions:

• Object classification directly from MS-STIL echo data is a viable alternative to using reconstructed depth images.
• The deep learning approach shows promise for improving MS-STIL system identification efficiency.
• Further dataset expansion and model optimization are recommended for handling a larger number of target classes.