Plant Classification Using Artificial Neural Networks

Automatic plant species identification is a difficulty challenge and an interesting area of research for both botanical taxonomy and computer science. From the past few years, some attempts towards the development of automatic plant recognition systems have been proposed, but the performance of such systems is not satisfactory in terms of accuracy, and these systems are also task dependent, since they are strongly influenced by the set of characteristics extracted from plant samples, leading to the problem known as data set bias. In this work, we use a Multi-Layer Perceptron (MLP) artificial neural network trained with Backpropagation algorithm to perform automatic plant classification. To avoid data set bias problem, some plant data sets which use different plant features obtained by different feature extraction processes are employed. We compare MLP algorithm with several supervised learning methods from plant recognition literature using a statistical hypothesis test of type Friedman/Nemenyi test. The obtained results show the potential of MLP algorithm to deal with plant classification in a unbiased fashion.