In order to count soybean seeds quickly and accurately, improve the speed of seed test and the level of soybean breeding , this dissertation developed a method of soybean seed counting based on VGG-Two (VGG-T). Firstly, in view of the lack of available image dataset in the field of soybean seed counting, a fast target point labeling method of combining pre-annotation based on digital image processing technology with manual correction annotation is proposed to speed up the establishment of publicly available soybean seed image dataset with annotation. This method only takes 197 minutes to mark 37,563 seeds, which saves 1,592 minutes than ordinary manual marking and replaces 96% of manual workload. Secondly, a method that would combine the density estimation-based methods and the convolution neural network (CNN)-based methods is developed to accurately estimate the seed count from an individual threshed seed image with a single perspective. Finally, the model is tested, and verify the effectiveness of the algorithm through three comparative experiments (with and without data enhancement, VGG16 and VGG-T, multiple sets of test set), which respectively provided 0.6 and 0.2 mean absolute error (MAE) in the original image and patch cases, while mean squared error (MSE) is 0.6 and 0.3. Compared with traditional image image morphology operations, ResNet18, ResNet18-T and vgg16, this method improves the accuracy of soybean seed counting.