In order to detect the odor of salmon refrigerated at different refrigerating temperatures and identify its freshness more accurately, an electronic nose based on kernelbased machine learning model was designed. It consisted of five parts, which were the detection air chamber, the array of six gas sensors, the data acquisition module, the pattern recognition module and the display interface. Kernelbased machine learning model was selected as the pattern recognition method of the electronic nose, and support vector machine (SVM) was selected as the learning machine of kernelbased machine learning model. The odor fingerprint data of salmon samples respectively refrigerated at 0℃, 4℃ and 6℃ was collected to train and test the kernelbased machine learning models with different kernel functions and kernel parameters. Finally, a kernelbased machine learning model that had the best salmon freshness identification effect was determined. And it was determined that the polynomial function was taken in the kernel function, and the kernel parameters of q, γ and c were taken as 3, 15 and 0, respectively. Analysis of identification result of test set salmon samples was conducted, which showed that no days deviation correct rate was 57.14% and allowable deviation of 1 day correct rate was 92.86% at 0℃, no days deviation correct rate was 88.89% and allowable deviation of 1 day correct rate was 100% at 4℃, no days deviation correct rate was 75.00% and allowable deviation of 1 day correct rate was 100% at 6℃. It proved that the model had certain ability to identify the freshness of salmons refrigerated at different temperatures. Compared with the result of principal component analysis (PCA), the kernelbased machine learning model had a better ability.