The invention relates to a method for judging rotten blueberries based on a low-field nuclear magnetic resonance technology, and belongs to the field of agricultural product detection. The method comprises: firstly, collecting relaxation spectrum information of blueberries to be detected through a low-field nuclear magnetic resonance system, obtaining transverse relaxation time and signal amplitudes, and obtaining 6 transverse relaxation signal variables in total; collecting a proton density weighted image of the cross section of the blueberry to be detected by using a low-field nuclear magnetic resonance system to obtain five characteristic variables, namely a mean value, a gray scale number, correlation, inertia and a gray scale average; taking the six transverse relaxation signal variables and the five characteristic variables as input vectors, inputting the input vectors into a BPNN model, and judging rotten blueberries. According to the method for judging the rotten blueberries, when the rotten blueberries are detected through the low-field nuclear magnetic resonance technology, the method is not influenced by unobvious rotting characteristic and placement directions of the blueberries with the deep color, multi-layer imaging can be further conducted on the interiors of the blueberries, and internal tissue changes caused in the blueberry rotting process are explored.