Articles
NEURAL NETWORKS FOR THE CLASSIFICATION OF POT PLANTS
Article number
421_2
Pages
37 – 48
Language
Abstract
Full grown Spathiphyllum plants are classified by human, by a neural classification system (NCS) and by means of multiple regression (MLR). The inputs of the NCS and the MLR systems are plant features, calculated from images, acquired by digital image processing.
The results of the NCS, a fully connected feed forward system, and the MLR classifications are compared with the human classification.
The human grader is handled as an expert and the classifications of the NCS and MLR models ideally are identical.
However the human grader is the expert, to a certain degree he is inconsistent in itself.
Unless these unavoidable inconsistencies his errorneous classifications of the pot plants are used to train a neural network.
To reduce the influence of expert errors on the modelling of the neural system global optimisation procedures are used in stead of the common used method : backpropagation.
The adaptive properties of neural systems are exchanged for a better handling of errorneous inputs.
The results of the NCS, a fully connected feed forward system, and the MLR classifications are compared with the human classification.
The human grader is handled as an expert and the classifications of the NCS and MLR models ideally are identical.
However the human grader is the expert, to a certain degree he is inconsistent in itself.
Unless these unavoidable inconsistencies his errorneous classifications of the pot plants are used to train a neural network.
To reduce the influence of expert errors on the modelling of the neural system global optimisation procedures are used in stead of the common used method : backpropagation.
The adaptive properties of neural systems are exchanged for a better handling of errorneous inputs.
Authors
J. Meuleman, J. Dijkstra
Keywords
Online Articles (31)