A method of extracting watercore feature and calculating proportion based on BiSeNet was proposed to solve the problems of low precision and complex methods of evaluating watercore quality during the planting process due to the irregular distribution of watercore characteristics in the cross section of watercore apple. Four models including BiSeNet， FCN， PPLiteSeg and DeepLabV3 were used to extract the watercore characteristics in the cross section of 380 watercore apple samples. Each evaluation index in theoretical research and practical research were calculated separately. The comprehensive evaluation and comparison was conducted. The results showed that BiSeNet was superior to the other three network models in both training time and training accuracy. It took 130 s， with the accuracy rate of 98.36% and the intersection and combination ratio of 85.1%. In the actual proportion calculation， the average proportion calculation error was 4.04%， lower than that of the other three models， and there was no large deviation. It is indicated that the method of extracting and calculating watercore characteristics in the cross section of watercore apple based on BiSeNet can provide more accurate methods of evaluation and comparison targets for nondestructive testing of watercore characteristics of watercore apples while providing specific proportion of watercore.