Transmission Electron Microscopy (TEM) provides to display particles at nano level. Detection, classification, localization, and counting these particles on images at different scales are difficult processes because of the overlapping particles, different sizes, and densities. In this paper, an automatic detection of magnetite (Fe3O4) particles on TEM images based on Convolutional Neural Networks (CNN) is proposed. First, CNN is used to distinguish particles from background. The windows classified as particles are re-evaluated for classification of a pixel as inside, border, and outside. Using the results of the final classification, the center and radius features of detected particle boundaries are determined by Hough transform. The proposed method is tested on 13 different images and produced successful results.