Vaginal cleanliness degree is one of the most important diagnosis criteria in regular examination of leucorrhea. It can be used to judge whether inflammation occurs. The most common clinical way to obtain vaginal cleanliness degree is manual microscopic examination. In current research, some groups have already proposed some methods to automatically analyse images from microscopic examination of leucorrhea secretion. But none of these methods can recognize four important targets at the same time, so they can’t obtain the vaginal cleanliness degree. This paper reports about a method that can automatically calculate the vaginal cleanliness degree of a microscopic examination image. This method includes a convolutional network similar to fully convolutional networks, and some morphological operations to extract contour of each cell and count their number. Experiments prove that our algorithm is simple, fast, efficient and accurate. It has good clinical potential.