In order to solve the problems in the process of feeding and management of meat rabbits, such as stress caused by manual weighing and difficulty in process of collecting weight information, a method of image segmentation and weight estimation based on deep convolution neural network was proposed, which can realize the contactless weighing of rabbits. A rabbit instance segmentation network based on Mask R-CNN was constructed. The residual network ResNet101 was used as the backbone network, and COCO dataset was used for migration learning to improve the training efficiency and obtain the segmentation results of unrestricted meat rabbits in the fence. Then the pixel area of each sample mask was extracted, and curvature and body length were introduced to modify the weight relationship between each sample and the corresponding weight. Projection area, curvature, body length and age as input parameters and body weight as output parameters, a six neuron weight estimation neural network was constructed to test the rabbit instance segmentation network and weight estimation neural network, the results showed that when IoU was 0.5∶0.95, the classification accuracy of rabbit segmentation network was 94.5%, and the pixel segmentation accuracy was 95.1%. The fitting correlation coefficient R of weight estimation neural network was 0.99391, MSE was 0.0336, and the mean weight error was 123g. The model had a good prediction effect on meat rabbits of different ages and different postures.