Background

Image classification is a widely used tool in today's world to further enhance our society. Image classification involves the use of images and labels to help train a model to be able to identify new images, and then put them into predefined categories. Applications of such a tool can involve autonomous driving systems, quality control, and medical image analysis. Specifically, the Navy is interested in the applications pertaining to the defense sector such as a target identification and threat assessment. In the Navy, being able to classify images can be leveraged in a variety of different applications. Having the ability to identify aircrafts can allow agents to accomplish various tasks such as identify if a platform is military and its origin. Identification mistakes can have serious implications to the warfighter; therefore, creating image classifiers that identify aircraft are a crucial component of artificial intelligence (AI) research to which the Navy could benefit.

Challenge Overview

Objective: Create and fine-tune an image classification model that can detect and identify images from a given dataset. The goal of the project is to demonstrate key concepts such as data preprocessing, model architect design, and training techniques that can be transferable to military applications that involve the need for image classification.

Challenge: Students will take a Military Aircraft Detection Dataset and create an Image Classifier that will be able to detect and identify an aircraft. This will be achieved through means of preprocessing the provided data and then creating an architecture that can be trained with that data. Students should look to create a model with high accuracy to keep up with the demand for improved image classification requirements for military applications.

Dataset: The dataset that will be used is open sourced and can be found on Kaggle.com.

• The Military Aircraft Detection Dataset is the dataset that will be used to train the models and consists of a myriad of military aircrafts. Military Aircraft Detection Dataset (kaggle.com)

Guiding Questions

• What kind of data preparation can be done to help improve the model? How can the data be split into training, validation, and test sets?
  
  
• What image augmentation can be done to improve accuracy?
  
  
• Which deep learning model is most suitable for image classification?