To address the limited imaging resolution and depth-quantification accuracy in the detection of hidden defects within the rail head, an ultrasonic phased-array detection and imaging-optimization procedure for rail-head defects is proposed. A two-dimensional acoustic-field model of the rail head is established in COMSOL to investigate the influence of array parameters on the focused beam distribution and to optimize the probe-parameter set. Inspection data are acquired using Full Matrix Capture (FMC) and reconstructed by the Total Focusing Method (TFM), and an imaging-optimization strategy combining normalization-based denoising, the Hilbert transform, and TF-PCF weighting is introduced. Experiments are conducted on a P60 rail test block with prefabricated flat-bottom-hole defects in the rail head, and the results are validated by comparison with simulations. The results show that improved imaging performance is obtained with 32 elements, an element-pitch-to-wavelength ratio of 0.5, an element-width-to-pitch ratio below 0.5, and a steering angle not exceeding 45°. The optimized imaging suppresses noise and artifacts and enables defect localization and quantitative measurement, with a hole-length error within 10% and a depth error within 5%. This study provides a theoretical basis for high-resolution imaging and quantitative evaluation of rail-head defects in ultrasonic phased-array inspection.