Techniques for milled rice kernel detection are crucial for ensuring quality from adulterated grains where a machine learning-based approach is desirable. This study proposes an efficient system for milled rice kernel classification and counts the number of rice kernels in an image to find the percent of adulteration. It has different phases including image processing, image segmentation, image identification, and counting the number of rice kernels. A CNN model consisting of 13 input layers, 6 hidden layers, and an output layer is applied to classify rice grains. A self-generated dataset is developed by taking images of three different types of grains, e.g., Basmati, Birui, and Atop. The performance of the method is compared with two different deep learning algorithms and the classification accuracy as follows CNN, YOLOv8, and GoogleNet methods are 99.39%, 95.71%, and 99.10% respectively. This is the very first attempt to develop an artificial intelligence system for identifying adulterated milled rice kernels in Bangladesh varieties. Therefore this model can be a good option for the rice milling industry to classify the appropriate rice type and ensure a premium quality of rice.