Determination of Cuscuta Campestris L. inflection seed population by using digital image analysis and artificial neural network

Cuscuta Campestris L. is one of the alfalfa weeds. The best way to fight this weed is preventing field contamination and using Cuscuta free seeds. Now in many research centers¡ alfalfa seed infection percentage is manually calculated using of Binocular that requires high cost-time and has low accuracy. In this study¡ a computerized method was designed for calculating the percentage Cuscuta mixed with alfalfa seed. This method is based on image processing and application of artificial neural networks using a Matlab programming. Therefore¡ the desired images were obtained and then using a suitable threshold in Matlab programming¡ seeds were separated from the substrate surface. By using the geometrical relationship¡ dimensional characteristics of seeds were extracted. For both alfalfa seed and Cuscuta seed were used as inputs to the neural networks. To select the most suitable neural network topology¡ different types of networks with different transfer functions and learning functions and different numbers of neurons were examined and it was found that the topology 4-5-1whit Logsig transfer function and GDX learning function is able to detect the percentage of Cuscuta seed with a R2 of 0.956 and RMSE of 0.017.