Swirl-assisted distributed combustion was examined using a deep-learning framework. High intensity distributed combustion was fostered from a 5.72 MW/m3 atm thermal intensity swirl combustor (with methane fuel at equivalence ratio 0.9) by diluting the flowfield with carbon dioxide. Dilution of the flowfield caused reduction of global oxygen (%) content of the inlet mixture from 21% to 16% (in distributed combustion). The adiabatic flame temperature gradually reduced, resulting in decreased flame luminosity and increased flame thermal field uniformity. Gradual reduction of flame chemiluminescence was captured using high-speed imaging without any spectral filtering at different oxygen concentration (%) levels to gather the data input. Convolutional neural network (CNN) was developed from these images (with 85% of total data used for training and 15% for testing) for flames at O2 = 16%, 18%, 19%, and 21%. Hyperparameters were varied to optimize the model. New flame images at O2 = 20% and 17% were introduced to verify the image recognition capability of the trained model in terms of training image data. The results showed good promise of developed deep-learning-based convolutional neural network or machine learning neural network for efficient and effective recognition of the distributed combustion regime.