Defect apple recognition based on spectral imaging and improved ant lion optimized regularization extreme learning machine

• Defect apple recognition based on multi-spectral and improved IALO-RELM is proposed • The apple recognition performance of each band of the multispectral was analyzed • Improved IALO has better optimization performance on high-dimensional problems • The IALO-RELM classification model can optimally identify defective apples To address the problem of poor accuracy in identifying defective apples using traditional image-based methods, we propose a method for identifying damaged apples that combines multispectral imaging with an Improved Ant Lion Optimization-based Regularized Extreme Learning Machine (IALO-RELM). An apple multi-spectral data acquisition system was set up in the laboratory, and we selected the bands with high differences between defects and normal apples through experimental research. The improved ALO algorithm employs the CEC2005 benchmark function to verify the effectiveness of the improvement, and the results show that the improved ALO has better global search ability and stability. Then, we introduce the Softmax function into RELM to construct the classification model, and use the proposed improved IALO to optimize the weights and biases of RELM to identify defective apples. IALO-RELM has the best defect apple recognition performance using spectral data reduced by principal component analysis (PCA), with precision, recall, and F1 scores of 0.9909, 0.9884, and 0. 9896, respectively, which are higher than those of SVM, AdaBoost, and CART. It is observed that PCA effectively reduces the data dimensionality, thereby enhancing classification accuracy in apple multispectral data processing compared to Local Binary Pattern and Stochastic Neighbor Embedding.