Feature-based detection using Bayesian data fusion

Current cocaine detection techniques used at borders have their challenges, which include cost of training specialised operators, the high chance of operator error and the dangers involved in exposure of both operators and container contents to radioactive material. This paper describes a technique which utilises the benefits of data fusion to develop a non-invasive system which relies less on the expertise of the operator, whilst improving false positive rates. To improve the capabilities of the cocaine-detecting fibre-optic sensor, the raw data was pre-processed and features were identified and extracted. The output of each feature is a decision on the classification and the conditional probability that it belongs to the chosen class based on the observed data, which serve as input into a Bayesian data fusion module and outputs the probability that a sample belongs to a class based on the observed features and makes a decision based on the class with the higher probability. The results show that the Bayesian fusion module greatly improves the detection rates of individual feature.