Fraud detection of power meter reading using image matching with inception v3 convolutional neural network

Abstract

Corruption occurs everywhere, and even meter readers can be corrupted. One of the examples is the consumer that is unwilling to pay utility bill will collude with meter reader to manipulate the meter reading for the utilities. For preventing such a thing from happening the utilities company had introduced a method to the meter reader which tells them to attach the meter reading (number) along with the photo of the meter with the reading. The photo will be sent to the company to ensure the meter reading is correct and the number is not created or generated by the meter reader. There is nothing perfect in this world, with this prevention the meter reader is still able to find a loop hole that allows them to manipulate the meter reading. What they do is get the meter reading from another meter and use the number and photo for the other consumer. This causes a lot of issues to the customer. Where the victim will need to pay more than the usual caused by the image of the meter is swapped. In this project, a solution is introduced that is able to reduce fraud from the meter reader by using image matching in fraud detection. This can also reduce the dependency on human checking of the image from the meter reader for the meter reading fraud detection. In this project, image matching algorithm is introduced to match the new meter image with the image from the image database, the image matching algorithm will also inform if any fraud is found from the image through Google Inception V3 as a proposed model with the accuracy of 99.1285%.