DEVELOPMENT OF A NEURAL NETWORK MODEL FOR IDENTIFYING BULK COWPEA SEEDS VARIETY USING ITS ELECTRICAL PROPERTIES
Artificial intelligence using machine leaning algorithms are modern trends in global industrialization. For agriculture to meet the global demand, the need to automate it processes are crucial. The objective of this study was to develop an artificial neural network model; that will be used to detect and identify variety of cowpea seeds in large storage facilities, using its electrical properties. Electrical properties of three variety of cowpea were generated; at five different moisture content, with five different current frequencies. A three-layer model was developed using multi-layer Perceptron method. It was trained and optimized using batch and scaled conjugate gradient methods respectively. Activation functions used were hyperbolic tangent and Softmax for the hidden and output layers; covariates in the input layer were standardized. The developed network model identifies 96, 97 and 93% varieties correctly during training, testing and validation respectively. Receiver Operating Characteristics (ROC) curve plotted for the model performance shows areas under the curve to be above 0.9 for all variety identified. This shows that the model performance was over 90% for predicting all varieties. The cumulative gain and lift charts were plotted to evaluate the model. Inductance was diagnosed to be the most important predictor to the model, while current frequency was the least. Pair t – test analysis at p<0.01, was done to further validate the model. This developed artificial neural network model can be used to program electrical sensors to identify cowpea seeds varieties during bulk storage, handling and processing. Such device can be used for quality control.