End to End Machine Learning Pipeline

Problem Statement

Currently, the autonomous inspection program that the vehicle rely heavily on computer vision models to ensure that the inspection is CSWIP 3.4U compliant. However, upon entering the new location, the computer vision model fails since the location is new, where training data cannot be captured beforehand due to deployment and security constraints. New data can only be collected during deployment which are short intervals and the personnel can only collect data that are related to the job during manual inspection. The amount of data collected also cannot be large due to the weak WIFI connections on-site.

Solution

An end-to-end machine learning pipeline was built to streamline the process of training a new model. Multiple machine learning methods are implemented to reduce the number of images used for retraining along with analysis of the performance of the model on dataset to ensure that the model can perform well during deployment.