The invention provides a cervical cancer lesion diagnosis method fusing multi-modal prior pathological depth features in the field of medical image processing. The method comprises the following steps: step S10, acquiring a cervical image, pathological definite diagnosis data and annotation information; s20, inputting the cervix uteri image and the annotation information into a deep neural networkmodel for training, and generating a first-stage training result; s30, based on the pathology definite diagnosis data and the first-stage training result, coding the cervix uteri image by adopting asmall network, performing second-stage fusion on the first-stage training result, inputting the first-stage training result into a deep neural network model for training, and generating a second-stagetraining result; s40, determining backbone network parameters, inputting the backbone network parameters into the deep neural network model to perform progressive migration training on the cervical image, and generating a three-stage training result; and S50, carrying out diagnosis classification on the cervical images by utilizing a three-stage training result. The method has the advantage thatthe accuracy and efficiency of cervical cancer lesion diagnosis are greatly improved.