Abstract: In order to solve the problems of different body weight of sows, low reproductive capacity of sows and high culling rate of piglets caused by the difficulty in accurately controlling the feeding amount of pregnant sows, a feeding station for gestation sows based on machine vision was designed in this experiment. The feeding station realized the accurate feeding of gestation sows by adopting machine vision image processing technology, RFID（radio frequency indentification） technology, STM32 microcontroller, industrial camera, communication technology, software technology, database technology and independent feeding mechanical structure. The RFID technology was used to identify individual sows through the electronic ear tags worn by sows. The STM32 microcontroller was used as the embedded control unit to communicate with the RFID reader, which triggered the camera to take pictures of the sows entering the station and realized the acquisition of sow condition data through the image processing technology. The feeding quantity was calculated based on the corresponding parameters such as sow condition, litter size and gestation time, so as to control the feeding motor and realize accurate feeding. The feeding accuracy was calculated by counting the feeding data of multiple groups from six feeding stations and comparing and analyzing the difference between the theoretical feeding amount and the actual feeding amount. After that, the accuracy of body condition classification of feeding station was calculated by manual backfat measurement and feeding station measurement of 70 pregnant sows. The results showed that the feeding accuracy of the feeding station was higher than 95%, which could realize accurate feeding, and the accuracy rate of body condition grade assessment reached 92.9%, which could automatically obtain the body condition grade of gestation sows, and the system ran stably. The results indicated that the feeding station of gestation sows based on machine vision could automatically obtain the body condition grade of gestation sows, and achieve accurate feeding of sows based on relevant parameters, reduce breeding costs, improve economic efficiency and realize intelligent breeding of pigs.