Wheat kernels, as a type of living organisms, will continue consuming the nutrients of themselves to maintain their vital activities during the normal storage period. With the increase of storage time, various enzymes inside wheat kernels decrease or lose their activities, the intensity of respiration decreasing gradually, the colloid structure of protoplasm getting relaxed, and then the physical and chemical states of wheat kernels have changed, which result in the deterioration of subsequent edible and processing quality. Therefore, it is of great economic value and social significance for our country to carry out accurate fresh degree detection to stored wheat and ensure the quantity and quality of wheat kernels. The identification methods commonly used for fresh wheat degree mainly include sensory determination method and various biochemical methods. The former method with poor repeatability, mainly depending on the operator's subjective experiences, is easily disturbed by external factors, and has an obvious error in the determination results, which is always used as a sort of auxiliary testing method in the aspect of wheat quality detection. Although the latter's accuracy is high, the whole detection process is time-consuming, and it is usually involved in a complex pretreatment for the tested samples. Meanwhile, various chemical reagents used in the detection process may cause certain pollution to the environment. Thus, it is urgent to establish a fast, accurate and green identification method for the fresh wheat degree. Special biophotonic instruments have tested biophoton signals of stored wheat kernels in five different years in this paper, and then combined the improved multiscale permutation entropy algorithm to analyze the features of wheat biophoton signals in four years from 2015 to 2018, finally, taking advantage of backpropagation neural network to classify the fresh wheat degree in four years. Experimental results show that there exist certain differences in the spontaneous biophoton number of wheat kernels stored in different years, among of which the biophoton numbers of wheat kernels in 2019 are much larger than the numbers in the other years, and the permutation entropy value of biophoton numbers of the rest wheat samples shows an increasing trend with the extension of storage time. It has been validated by simulation experiment that the improved algorithm greatly solves the problems of signal dither and mutation that existed in the MPE algorithm which can be used as an obvious feature to characterize the fresh degree of wheat kernels. After simulating by backpropagation neural network, the recognition accuracy rate of the novel classification model proposed in this paper can reach 95% and be able to precisely determine the fresh degree of wheat kernels.