The detection of ship trail clouds by artificial neural network
The work described in this paper is concerned with the analysis of AVHRR data which visually exhibit ship trail clouds. The aim was to devise an automatic means of identifying these features. The existence of ship trail clouds has been known for some time, though the exact mechanism for their formation is not yet fully understood. They are normally associated with low-level stratus and stratocumulus cloud and are thought to be caused by the presence of condensation nuclei emanating from the ship exhaust funnel. Given the right conditions, these trails can be easily seen in both visible and infrared imagery and under calm conditions they can exist for several days. Edge-detection analysis was found to be unsatisfactory for identification purposes and hence a neural approach was adopted. The neural analysis was shown to be very promising. Application of a neural network trained on four images resulted in the detection of the presence of most of the trails in the dataset on which it was trained. Just as importantly, the network did not, on the whole, confuse cloud with ship trails.