Dental caries is a highly prevalent oral disorder, and the strategies of deep learning have been employed for diagnosing caries with large populations by employing RGB images. The conventional attention-aided image categorization approaches have the issues of feature underutilization and simple interference by irrelevant and background data. The timely recognition of dental caries is significant for performing treatments. For this objective, the bitewing radiography is employed to provide the initial detection of caries. The utilization of deep structured models with neural network approaches helps process the large number of images that have been experimented with nowadays and provides promising functionalities. The computer-assisted smart vision approaches applied by image processing and machine learning mechanisms are required to eliminate these drawbacks. Hence, deep learning mechanisms have accomplished remarkable diagnosis efficacy in the radiology sector. Hence, a novel deep learning-based dental caries detection and classification framework is designed in this work. Initially, Cone Beam Computed Tomography (CBCT) images utilized for the detection and classification of dental caries are collected from the benchmark resources. Further, the collected images are provided for the image segmentation phase. Here, the developed Nested Dilated-based Transformer Recurrent Residual Unet (NDT-R2Unet) framework is used to attain the segmented images, and these images are provided as the input to the dental caries disease classification phase. In this phase, an Adaptive Efficient Net with Region Attention (AdaENet-RA) is employed to classify dental caries among the individuals. Moreover, several parameters in developed AdaENet-RA are tuned using Arbitrary Updated Wild Geese Migration Optimization (AUWGO) and provide the dental caries disease classified outcomes. Later, various validations are executed in the developed framework to verify the effectualness of the suggested framework over the classical techniques.