Comparison of the performance of ANN and ANFIS models in acoustic detection and classification of different almond varieties

Due to the possibility of mixing hard, semi soft and soft almonds with different market value after harvesting, using an effective separation system for supplying uniform products to the market is essential. In this research, in order to classify almond varieties, an intelligent impact-acoustic system was used. The system operation was done by dropping almond nuts onto a steel impact plate through a pipe. Then some features such as amplitude, phase and power spectral density (PSD) of almond nuts were extracted from the analysis of sound signal gained by a microphone in both time and frequency domains by means of Fast Fourier Transform (FFT). Principal component analysis method was used for reduction of the features. Two types of artificial intelligence techniques including Artificial Neural Networks (ANNs) and Adaptive Neuro-Fuzzy Inference System (ANFIS) were used to classify almond nuts and their performances were then compared. The ANN model used the multi-layer perceptron network with back propagation algorithm and Levenberg–Marquardt algorithm (LM) learning function. In ANFIS model, due to the limitation of inputs number, three principal components of PSD feature that had higher priority were selected as inputs and the almond classes as outputs. Also, hybrid optimization techniques were used for classification. By comparing the artificial intelligence techniques used here, ANN model with about 96.2% accuracy had better performance to classify almond nuts than that of the ANFIS model with 81% accuracy.