Tunicate Henry gas solubility optimization‐based deep residual network for fruit ripeness classification

Abstract

Summary Fruits are considered a significant component to offer humans adequate nutrients, like carbohydrates, vitamins, and dietary fiber but the ripening of fruits is a major issue as the ripening of fruits is detected considering more or less time based on microclimate conditions. This article proposes a new technique to classify fruit ripening using the deep model. Here, the input images like mango, apple, and banana are fed to preprocessing phase wherein Gaussian filtering is adapted to eliminate unwanted distortions. Features like color histogram, histogram of gradients, statistical features, local Gabor binary patterns, significant local binary pattern, and convolution neural network (CNN) features are mined to classify fruit. The DCNN training is done with the adopted tunicate‐based Henry gas solubility optimization algorithm (THGSO) obtained by integrating the tunicate swarm algorithm and Henry gas solubility optimization (HGSO). Then, ripening classification is done with deep residual network (DRN). The DRN training is done using the adopted THGSO. The adopted THGSO‐based DRN presented developed fruit classification performance with elevated accuracy 91.4%, sensitivity 92.5%, and specificity 87.5%. The adopted THGSO‐based DRN provided improved fruit ripening performance with maximum accuracy 92.5%, sensitivity 93.5%, and specificity 90.5%.