Dry fruit image classification using stacking ensemble model

Precise and efficient classification of dry fruit images is critical for enhancing quality control, efficiency, and safety in the agricultural and food industries. This study presents a CNN-based classification model developed to analyze a diverse dataset of dry fruit images. The dataset comprises 11,520 high-resolution images representing four varieties (raisins, cashews, almonds, and figs), each divided into three groups, resulting in twelve distinct classes. The proposed multi-step methodology covers data collection, preprocessing, augmentation, and model training. A wide range of image conditions, including variations in fruit types, shapes, colors, and lighting, facilitated comprehensive experimentation and validation using performance indicators such as accuracy. A stacking ensemble, integrating predictions from multiple models (e.g., VGG16, VGG19, ResNet50, MobileNetV1, Mo- bileNetV2, SqueezeNet, and ShuffleNet), achieved a test accuracy of 98.32 %, surpassing individual base model performances (MobileNetV2: 90.33 %, Mo- bileNetV1: 93.68 %, SqueezeNet: 96.98 %, and ShuffleNetV2: 97.79 %). These findings underscore the model’s potential for real-time applications in qual-ity control, automated processing, nutritional research, and other domains, while also delineating the key components required for accurate classification.