The extension of the Standard Model (SM) field content with one inert Higgs doublet (IHD) and three right-handed neutrinos (RHNs) is a well-motivated approach. The key advantages of the model include the appearance of a weakly interacting massive particle (WIMP) like dark matter (DM) candidate from the neutral component of the IHD, along with the plausible explanation of the sub-eV mass range of SM neutrinos via the radiative seesaw mechanism. Additionally, the decay of RHNs can contextualize the baryon asymmetry of the universe via leptogenesis and is intricately connected to CP violation. Also, given the ongoing searches for light scalars at various experimental facilities, the extended Higgs sector of the model continues to be at the forefront. However, this scotogenic framework encounters a deficiency in providing the observed amount of relic density for a particular mass range $\sim (80 - 500)$ GeV of its DM candidate, hence requiring further augmentation. Also, the WIMP scenarios have not yet resulted in conclusive hints at the direct detection experiments. In this context, our work is based on further extension of the above Scotogenic model by a dark sector. Additionally, considering the cosmic coincidence aspect, we operate within the framework of two-sector leptogenesis. To have a predictive flavor structure in the visible sector, we impose $A_4$ symmetry. Also, we adhere to spontaneous CP violation via complex vacuum expectation value of the falvon field, leading to a situation where there is only one CP-violating phase as a common connection between the visible and dark sectors. In our analysis, we find for the lightest RHN mass $\sim 10^{10}$ GeV, our results are in good agreement with the observational ratio of relic densities, i.e., $\Omega_{\rm DM}/\Omega_{\rm b} \sim 5$ for a few GeV range of mass of the dark sector DM candidate.

Hyperspectral image (HSI) classification faces critical challenges, including high spectral dimensionality, complex spectral-spatial correlations, and limited training samples with severe class imbalance. While CNNs excel at local feature extraction and transformers capture long-range dependencies, their isolated application yields suboptimal results due to quadratic complexity and insufficient inductive biases. We propose CLAReSNet (Convolutional Latent Attention Residual Spectral Network), a hybrid architecture that integrates multi-scale convolutional extraction with transformer-style attention via an adaptive latent bottleneck. The model employs a multi-scale convolutional stem with deep residual blocks and an enhanced Convolutional Block Attention Module for hierarchical spatial features, followed by spectral encoder layers combining bidirectional RNNs (LSTM/GRU) with Multi-Scale Spectral Latent Attention (MSLA). MSLA reduces complexity from $\mathcal{O}(T^2D)$ to $\mathcal{O}(T\log(T)D)$ by adaptive latent token allocation (8-64 tokens) that scales logarithmically with the sequence length. Hierarchical cross-attention fusion dynamically aggregates multi-level representations for robust classification. Experiments conducted on the Indian Pines and Salinas datasets show state-of-the-art performance, achieving overall accuracies of 99.71% and 99.96%, significantly surpassing HybridSN, SSRN, and SpectralFormer. The learned embeddings exhibit superior inter-class separability and compact intra-class clustering, validating CLAReSNet's effectiveness under limited samples and severe class imbalance.