Achieving consistent surface quality in direct laser interference patterning (DLIP) demands real-time insight into ultrafast laser–material interactions, particularly when structuring complex alloys such as Ti-6Al-4V. This work presents a hybrid image-to-signal machine learning framework that links offline topography characterization with real-time sensor data to enable predictive surface quality assessment. Periodic microstructures are fabricated using a picosecond pulsed laser equipped with a two-beam interference head and an off-axis photodiode for in situ optical monitoring. Ground-truth labels are generated from white light interferometry (WLI) images processed in the frequency domain via a 2D fast Fourier transform to extract radial power spectral density profiles. Spectral entropy serves as a quantitative indicator of texture order and enables unsupervised KMeans clustering into acceptable and nonacceptable quality classes. These entropy-based labels are assigned to the corresponding time-resolved photodiode signals and laser parameters recorded during fabrication. A supervised 1D convolutional neural network (1D-CNN) is then trained to predict surface quality using only the sensor data and process inputs. The model achieves a classification accuracy of 90%, demonstrating reliable detection of structural deviations without post-process metrology. This entropy-informed, sensor-driven framework highlights the potential of machine learning for real-time quality assurance in laser-based manufacturing systems.