Autophagy is an essential process in which cells degrade and recycle its damaged organelles and proteins to maintain cellular homeostasis and cope with stress. Based upon the publicly available CELLULAR dataset, images of Drosophila melanogaster S2 cells, two curated subsets of 250 and 1,150 high resolution images have been annotated based on the fluorescence and augmented with thresholding using HSV method. The proposed system combines classical image processing techniques and convolutional neural networks to improve the detection of auto-phagosomes and auto-lysosomes. Preprocessing was carried out including resizing, normalization, contrast enhancement, and noise reduction to normalize the inputs. Segmentation was carried out with OpenCV thresholding and a U-Net model was trained over binary masks from the fluorescence masks. For classification, the custom convolutional neural network (CNN) and transfer architectures including MobileNetV2, DenseNet121, and ResNet18 were tested. The custom CNN structure was found to be the best with an accuracy of 96.89%, superior to other pre-trained models. Our work combines the conventional imaging techniques with deep learning algorithms to accomplish scalable and reproducible biomedical image analysis.