Detection and recognition of surface-laid mines

This report summarizes the signal processing work carried out within the project Multioptical mine detection (MOMS). A number of methods for and aspects of data registration, anomaly detection, feature extraction, data fusion and mine recognition are described and discussed. A number of especially interesting methods have been tested and evaluated with sensor data from different scenes, in order to allow for analysis of pros and cons under certain conditions. Conclusions from the experiments are presented and discussed, with focus on aspects concerning signal processing in a sensor system perspective. A number of electro-optical sensors, passive as well as active have been considered within MOMS. In this report, a method for optimized sensor design is presented, that provides a tool for designing a relatively simple sensor that still is adequate for the task. This can be achieved through analysis based on information theory, in which the spectral characteristics of the sensor are defined based on the information they contain. In order for data from several sensors to be combined, the data has to been registered, i.e., transformed into a common coordinate system. The quality of the registration strongly influences the level at which data can be combined; ideal registration allows for fusion on the lowest level (pixel- or signal-level). In a distributed sensor system where, say, data from an airborne system shall be combined with data from a groundbased sensor, pixel-level fusion will probably be difficult to use. From a signal processing perspective, it is desirable that the sensors are mounted close to each other, preferably with common optics and/or detector array, so that the registration can be as accurate as possible. Among the signal processing techniques considered, anomaly detection emerges as a key component in a system concept. This method detects things that are different from what is expected (the background) and thus gives a first indication of possible mines. In addition, this technique can potentially be used for detection of other objects, e.g. IED's. The detected anomalies are then analyzed further, by using certain assumptions concerning the targets, e.g. their expected size. This leads to the detection of mine-like objects. If available detailed data about certain targets, e.g. CAD models or images, given before or collected during the mission, can be used for mine recognition, in which the similarity between detected objects and these targets is investigated.