Neural networks were used to classify agricultural crops from weeds using spectral reflectance measured over the visible to near-IR range. The neural network interconnections (weights) were pruned to obtain a minimal sized network using a Algorithm of Neural Network Size Extension. Reflectance from common crops (Zea mays, Heliantus cultus) and weeds (Amaranthus retroflexus, Setaria viridis, Sonchus arvensis, Shenopodium album, Euphorbia falcata, Agropyrum repens) were used to train and test the neural networks. Results show that the neural networks were capable of predicting samples used in training at 100% accuracy, and at 98% accuracy for samples used in testing.
Weitere Kapitel dieses Buchs durch Wischen aufrufen
- Neural Network Method in Plant Spectral Recognition
- Springer Netherlands
- Chapter 7