Selecting optimal wavelengths for detection of insect infested tomatoes based on SIMCA-aided CFS algorithm

• The optimal wavelengths for detection of insect infestation in tomatoes are selected.
• Soft independent modeling of class analogy is able to extract the effective wavebands.
• Correlation-based feature selection algorithm is able to find the best features.
• Artificial neural network based on spectral difference features results in best accuracy.

Selection of effective wavelengths is a decisive step for online multispectral imaging systems. In this study, a new approach was utilized to distinguish the most informative wavelengths for detection of insect infestation in tomatoes within 400–1100 nm. Soft independent modeling of class analogy (SIMCA) was first conducted in the entire spectral region after applying different pretreatment procedures. Following satisfactory results obtained from 1st derivative preprocessing (accuracy of 90%), the most effective wavebands for detection of infestation were attained by discrimination power plot of SIMCA analysis. Transmission differences between all possible pairs of wavelengths (T(λ1)–T(λ2)) in the obtained informative wavebands were then calculated to substitute the 1st derivative spectra. Afterward, correlation-based feature selection (CFS) algorithm was used to find the best pairs of wavelengths. To compare the performance of SIMCA-aided CFS procedure, CFS was also conducted on the raw spectral data. Seven spectral difference features and six wavelength features were found superior by CFS. To classify tomatoes, three different machine learning techniques including Bayesian networks (BNs), artificial neural networks (ANNs), and support vector machines (SVMs) were implemented. The test set validation results of all machine learning techniques revealed that the spectral difference features outperformed the raw spectra features, indicating the superiority of SIMCA-aided CFS procedure for detection of optimal wavelengths. Among different machine learning techniques, the best performance obtained by ANN based on spectral difference features with a classification accuracy of 95.0%. The results of this study can be adapted for developing an online tomato sorting system for detection of infestations.