The invention discloses a Polar code decoding method, a decoding device and a decoder. The decoding method comprises the following steps: acquiring modulated Polar code information at a channel receiving end, decoding a channel receiving value by using hard decision, and obtaining a preliminary decoding result, thereby obtaining a first estimated value of channel noises; using a convolutional neural network to establish a relationship between the first estimated value of the channel noises and a true value of the channel noises, and providing the first estimated value of the channel noises asan input to the convolutional neural network to obtain a second estimated value of the channel noises; correcting a Log likelihood Ratio LLR by using the channel receiving value and the distribution statistics of residual noises; and performing BP decoding iteration on the corrected LLR. According to the Polar code decoding method provided by the invention, the convolutional neural network is usedfor learning the characteristics of the channel noises and estimating the distribution of the channel noises, more reliable likelihood ratio information can be calculated to help the decoder to perform more accurate decoding, and the performance of a Polar code is improved.