The invention provides a fundus image retina arteriosclerosis detection method based on an improved encoding and decoding network, and the method comprises the steps: 1) collecting a fundus image, diagnosed as retina arteriosclerosis, of an ophthalmologist, fitting diseased blood vessels in the fundus image, and calculating and counting reflection parameters, so as to determine a retina arteriosclerosis detection threshold value; 2) improving a coding and decoding network by utilizing an Inception ResnetV2 module and a residual attention mechanism module, and applying the coding and decoding network to segmentation of artery blood vessels and artery reflective tapes; and 3) screening an effective area, sampling the effective area, fitting by utilizing a four-section Gaussian model to obtain a blood vessel gray scale distribution curve, calculating a reflection parameter according to a fitting result, comparing the reflection parameter with a threshold value, and judging whether the patient suffers from retinal arteriosclerosis or not. According to the method, the quantitative detection threshold of the retinal arteriosclerosis is determined, the deep learning technology is utilized, the problem that fundus arteriovenous blood vessels and artery reflective tapes cannot be accurately segmented through a traditional method is solved, and detection of the retinal arteriosclerosis is completed.