Pyrite framboids have widespread occurrence in sediments of all geological ages. The external spherical or spheroidal (ellipsoidal) form and the internal discrete and equant microcrystalline architecture are the two main characteristics of a framboidal pyrite. Framboid size distribution is fixed very early during diagenesis, and the framboids tend to be preserved through advanced stages of diagenesis. Size-distribution of framboids can be used to refer depositional redox conditions in both modern and ancient sediments. Sixty five scanning electron microscope (SEM) images at instrument magnification of 500X to 3200X from 14 locations along the burial depth of the Marcellus Shale were obtained. Around 3200 framboids were manually traced using JMicroVision to unravel the changes in water column oxygenation during the deposition of the Marcellus Shale. The objective of the study is to replace repetitive and time-consuming tasks with machines. Deep learning technique such as convolutional neural network (CNN) is a powerful approach for image analysis. We used Faster R-CNN Inception and ResNet architectures in this work. As a result, the CNN can detect up to ~ 99% of total framboids traced manually, and ~ 78% of framboids with 90% IoU score. Kolmogorov-Smirnov test was performed to estimate how extracted equivalent diameter framboid size distribution fit manually extracted data. P-values and corresponding D-statistic values show that distribution functions are comparable with insignificant deviations.