RPROP Algorithm in feature-level fusion recognition

RPROP is a fast BP algorithm in which the weights are adjusted local adaptively and the parameters can be chosen easily. The RPROP algorithm was introduced for feature-level fusion recognitions in this paper and the technologies of BP algorithms in feature-level recognitions were discussed here. The data of 5 vehicles were measured in darkroom with 2 different kinds of sensors, infrared radiation (IR) sensor and radar. Using features extracted from the data, the comparative experiments between the learning-rate descent BP (LDBP), the variable learning-rate BP (VLBP), the adaptive momentum BP (AMOBP) and resilient BP (RPROP) algorithms were emulated when they were applied in feature-level fusion recognition. The ROROP gives the most steady and fastest results even that high accuracy is demanded. All the algorithms had given the equivalent recognition rates after convergence at the equivalent error level. The experiments have proved that RPROP is an efficient BP algorithm for feature-level fusion recognition.