Accurately detecting and locating a large number of objects of interest in thin section images is an arduous task that requires assistance and is now possible thanks to the emergence of powerful deep learning techniques.
Our objective is to automate the detection and counting of small objects of interest, such as microfossils, in thin sections in order to facilitate and accelerate the analysis of samples by geologists.
We use a wide variety of state-of-the-art convolutional neural network algorithms. The annotation effort for the user can be greatly reduced by the use of our specific data pre-processing algorithms.
The resulting model is able to precisely locating the objects of interest on new images, enabling a direct and automated computation of occurrence statistics for the different classes of objects.
This method can be applied to 3D data from Computed Tomography for various image analysis applications. It is complementary with segmentation approaches for rock images. Besides it could integrate an automated fitting of elliptical shapes on the objects detected.