Image-free multi-character recognition.

Area of Science:

• Computer Vision
• Signal Processing
• Machine Learning

Background:

• Image-free sensing offers advantages in hardware and software by extracting semantic information directly from compressed measurements without image reconstruction.
• Existing image-free techniques struggle with classifying multi-semantic information from multiple targets within a practical field-of-view.

Purpose of the Study:

• To develop a novel image-free sensing technique capable of multi-target recognition.
• To address the limitations of current methods in handling complex, real-world scenarios with multiple objects.

Main Methods:

• A convolutional recurrent neural network (CRNN) architecture was employed, utilizing bidirectional Long Short-Term Memory (LSTM) layers.
• This CRNN framework was designed to predict the distribution of multiple characters simultaneously, capturing long-range dependencies in the data.

Main Results:

• The proposed CRNN-based image-free sensing technique demonstrated effectiveness in license plate detection.
• Achieved a recognition accuracy of 87.60% at a low sampling rate of 5%, with a high refresh rate exceeding 100 FPS.

Conclusions:

• The novel image-free sensing technique successfully tackles the challenge of multi-target recognition.
• The CRNN approach offers a promising solution for efficient and accurate object detection and recognition in resource-constrained environments.