Browsing by Author "Westlake, Samuel"

Now showing 1 - 4 of 4
  • Thumbnail Image
    ItemOpen Access
    Artificially Intelligent Targeting
    (Cranfield University, 2020-01-15 15:24) Westlake, Samuel
    The aim of this project is the development of new techniques for infrared anti-ship missile seekers. This image illustrates how we are using deep learning to detect, recognise and classify multiple ships. Our algorithm can differentiate between military and civilian vessels, and is even robust against the presence of infrared countermeasures and background clutter.In most cases, training deep learning algorithms requires thousands, if not millions, of carefully labelled examples. This presents a major challenge for the application of deep learning to infrared missile seekers, as the availability of such training data is extremely limited. To over come this, we simulated multiple thermal signatures for ten different ships and used these to synthetically generate a large and realistic data set. This data was then used to train our artificial neural network, and the subsequent model performed successfully on real-world infrared test data.
  • Thumbnail Image
    ItemOpen Access
    Deep Learning Techniques for Missile Seeker Automatic Target Recognition
    (Cranfield University, 2020-01-15 15:30) Westlake, Samuel
    Modern infrared missiles use sophisticated computer vision techniques, in conjunction with imaging seekers, to automatically detect and localise targets. However, simulations show that soft-kill countermeasures remain an effective defence against such systems. This research explores the feasibility of deep learning algorithms for Automatic Target Recognition (ATR) and aim to significantly improve seeker performance in the presence of soft-kill countermeasures and clutter. State-of-the-art neural network architectures were benchmarked using both simulated and real-world infrared data. Their performance was also analysed to inform tailored and novel developments. In addition to this improvement to existing capabilities, these algorithms established additional capabilities of target recognition and identification. This effectively enables target prioritisation and safeguarding of friendly assets
  • Thumbnail Image
    ItemOpen Access
    Deep Learning Techniques for Missile Seeker Automatic Target Recognition
    (Cranfield University, 2020-01-15 15:37) Westlake, Samuel
    Modern infrared missiles use sophisticated computer vision techniques, in conjunction with imaging seekers, to automatically detect and localise targets. However, simulations show that soft-kill countermeasures remain an effective defence against such systems. This research explores the feasibility of deep learning algorithms for Automatic Target Recognition (ATR) and aim to significantly improve seeker performance in the presence of soft-kill countermeasures and clutter. State-of-the-art neural network architectures were benchmarked using both simulated and real-world infrared data. Their performance was also analysed to inform tailored and novel developments. In addition to this improvement to existing capabilities, these algorithms established additional capabilities of target recognition and identification. This effectively enables target prioritisation and safeguarding of friendly assets.
  • Thumbnail Image
    ItemOpen Access
    IRShips
    (Cranfield University, 2020-11-13 16:07) Westlake, Samuel
    Welcome to the IRShips dataset of synthetically generated images for the development of IR anti-ship ATR algorithms. This dataset contains 972,000 images of 10 different ships, each with 9 different thermal appearances, taken from different ranges, bearings and elevations. Images are labelled with: ship class, ship type, bounding box coordinates, and much more metadata. See the README for more info.