Aiming at the trend of continuous expansion of fruit and tea gardens,the gradual development of intelligent agricultural mechanization and the lack of fruit and tea garden road scenes in common road semantic segmentation data sets,the semantic segmentation technology was used to some fruit and tea garden roads in Meizhou City,Guangdong Province to realize pixel-level segmentation of roads in fruit and tea gardens. Roads,people,and cars were used as classification objects to establish a scene image data set including 6 032 images of fruit and tea garden road. The data set was randomly divided into a training set including 5 429 images and a test set including 603 images according to a 9∶1 ratio. The MS-PSPNet semantic segmentation model was established based on the PSPNet (pyramid scene parsing network) segmentation model for optimization.The results of training showed that MS-PSPNet model mean intersection over union (mean intersection over union,MIoU) was 83.41%. The number of frames per second (frames per second,FPS) was 22.31. The MS-PSPNet model was applied to fruit and tea gardens under different road conditions and light intensity to conduct field tests and evaluate the accuracy. The results showed that the category pixel accuracy (mean pixel accuracy,MPA) of MS-PSPNet model exceeded 92%. MioU exceeded 91% in all cases of non-hardened road conditions. It is indicated that the MS-PSPNet model has good validity and applicability in road recognition of fruit and tea gardens.