Last Mile Delivery (LMD) relates to all activities in delivering the product to the end customer in the supply chain. Safe, secure, and on-time delivery of products to customers are the primary objectives of businesses rendering LMD services. LPG cylinder distribution in India is one such business that provides LMD to its subscribed customers through delivery agents employed by distributors. Despite being an essential commodity, there is no system to confirm or authenticate the delivery. This absence of the system required for authentication causes a rise in black-market deviation, negatively affecting the distributor's reliability and customer satisfaction. To the best of our knowledge, there is no research contribution made towards developing an authentication system for delivery confirmation in last-mile delivery of LPG cylinders. Against this backdrop, this study proposes a secure, reliable, cost-effective Delivery Authentication System (DAS) using artificial intelligence and image metadata. The proposed DAS has an image classifier module built using Convolution Neural Network (CNN) and a metadata processing module to extract geolocation. A sample image dataset of an LPG cylinder and objects which looks similar to an LPG cylinder (fire extinguisher, bottled water can, and cool drinks can) was used to train the image classifier. A prototype DAS with MobileNet V2 CNN architect is developed using Python and Keras library. With the data augmentation technique, we could train our CNN with a small dataset and prevent it from underfitting. The transfer learning technique helped our CNN to generalize well on the study data without overfitting and produced an accuracy score of 97%.  The expected outcome of this study is to reduce black-market deviation cases.