Intelligent Object Detector based on Deep Learning

To address the challenge of suboptimal performance in constant false alarm rate (CFAR) detectors due to statistical model mismatch, an intelligent detector known as RepVGG (Re-parameterization Visual Geometry Group) has been introduced. This detector leverages an end-To-end deep learning approach and is structured with RepVGG feature extraction modules and Rep VGG prediction output modules. A performance comparison between the designed RepVGG detector and the CFAR detector was conducted through Monte Carlo experiments. The results demonstrate that, under equivalent false alarm rate conditions, the Rep VGG detector consistently achieves higher detection probabilities than the CF AR detector when the signal-To-noise ratio falls within the range of 6dB to 15dB. The optimal performance is observed at signal-To-noise ratios of 9dB to 10dB, where it surpasses a 20% increase in detection probability.