Vaginal microflora pathological image is an important basis for the diagnosis of bacterial vaginosis, but analysis of the images manually takes a lot of time and effort, leading to low diagnosis efficiency, so new methods of automatic pathological image diagnosis need to be sought. In this paper, we proposed a model, ResLab, to diagnose vaginal microflora pathological image automatically. It took the pathological reports of gynecological examination as training set, and used deep learning technology to perform end-to-end analysis on the pathological images. The ResLab model predicted Nugent score to assist doctors in grading diagnosis. We optimized the ResLab in multiple ways to improve the prediction accuracy, by increasing the number of layers to extract deeper features, stacking two small convolution kernels to increase the receptive field, removing ReLU layers to reduce complexity, and replacing average pooling layer with max pooling layer to extract the most salient feature. It was proven that each optimization plan can significantly improve the perfomance of the model. The prediction accuracy of the ResLab model reached 82.19%, which outperformed VGG, GoogLeNet, ResNet. The ResLab model can provide doctors with relatively accurate reference results, thereby improving diagnosis efficiency and reducing diagnostic error.