Image noise plays a vital role in digital image processing. However, in some specific application scenarios, random noise has an uncontrollable effect on digital image processing. Besides, a large number of hyper parameters which need to be fine-tuned can lead to inefficient projects. Therefore, we propose a Image Noise Level Classification(INLC) technique for specific application scenarios by comparing image quality assessment(IQA) methods, fitting curves and designing two neural networks. For low-accuracy, we come up with a soft way by setting a tolerance rate to achieve a higher acceptable accuracy. Experiments show that our INLC is more accurate and efficient.