The aim of this project is to develop a generalized deep learning-based algorithm for processing stepwise signals generated from Single Molecule Technology (SMT) measurements, including single-molecule FRET, plasmonic probes, single molecule fluorescence, AFM, and optical/magnetic tweezers. The goal of this project is to minimize user subjectivity during the signal processing, including de-noising and state recognition, and overcome the challenges posed by the low signal-to-noise ratio (SNR) inherent in Single Molecule Technology data. Computational work over the next 12 months will focus on the following three main stages: Stage 1: Simulation dataset generation (2 months). We will generate datasets with diversity for training deep learning detectors. Noise with different powers will be added to Markov process in order to generate signals with different SNRs and to provide sufficient data with known noise characteristics for subsequent model training. Stage 2: Denoising Model Training (6 months). A diffusion model based on a U-Net architecture (DDPM) is planned to be trained using simulated data based on the generated artificial dataset to achieve effective denoising. Stage 3: Model Validation (4 months). The denoising methods such as common filters are tested and compared using real experimental data to confirm the denoising effect of the DDPM model.