Neural Network Model for the Evaluation of Lettuce Plant Growth

Abstract

A back propagation neural network was trained to evaluate lettuces in terms of plant growth characteristics, with a network consisting of 7, 8 and 5 processing units in the input, hidden and output layer, respectively. To generate the training data, clinorotation rates in the range between 0 and 25 rpm, centrifugation rates in the range between 0 and 5·5 rpm were selected for experiments to measure the daily plant width and height after transplant. Fifty-eight sets of training data were used. The training was terminated after 22 124 times of iterative calculations at the root mean square error value equal to 4·02×10-4. Ten sets of validation data were used to calculate the prediction error. The average prediction error was in the range between 2·5 and 9·7%. The ability of the neural network models to predict the required information is very accurate. As a result, there is a potential for the present technique to be applied to plant growth evaluating system under the simulated gravity conditions.