Enhanced convolutional neural network for apple shelf life prediction

Abstract The preservation of apples during storage is associated with the depletion of moisture and nutrients, which significantly impacts their shelf life. This paper introduces an enhanced convolutional neural network model for the recognition of apple shelf life. We have developed a methodology that involves capturing the acoustic and vibrational signatures of apples, followed by mapping their time-domain spectra through windowing, frame segmentation, and discrete Fourier transformation to construct Mel frequency spectrum diagrams. Subsequently, a substantial dataset of these spectral profiles is utilized to train a deep convolutional neural network, culminating in the generation of a predictive model. This model demonstrates a high degree of accuracy, with a test result accuracy of 95%, offering a valuable tool for the prediction of apple shelf life during storage.