Abstract: The 3D reconstruction of infected locust is very important for the biological popular science education. Especially, the vector contour extraction of infected locust slice image is the key step in aspects of illuminating the interactive process between the locust organ and bio-pesticide. Some classic segmentation algorithms aren’t suitable for the locust image with complex topology and minimal gray scale difference which will make the 3D reconstruction incomplete, inaccurate and non-vectorial. This study applies a new adaptive multiphase segmentation method of microscopic image based on improved Chan-Vese model to the 3D reconstruction. Firstly, in order to improve the speed and accuracy of contour extraction, the complex background is removed by scanning the pixel value in horizontal and vertical direction. And a great deal of isolated noises are reduced by the decision window. Secondly, the C-V model parameters λ1, λ2 and curvature are optimized by neglecting the curvature and setting the dynamical proportion of λ1and λ2. The exact extraction of tissue contour in the locust coelom has established a good foundation for the 3D reconstruction and organ deformation parameters calculation of coelom based on the above vector data.