Using Kolmogorov–Arnold network and ResNet for marine protein mapping in support of the Prabowo–Gibran MBG program

Accurate classification of marine species is critical for sustainable fisheries management, biodiversity conservation, and aquaculture optimization. This study leverages remote sensing data and a Kolmogorov–Arnold Networks-enhanced ResNet (ResNet-KAN) model to classify fish, shrimp, and seaweed distributions with high precision. A multi-spectral feature set, including sea surface temperature, chlorophyll concentration, oxygen availability, and salinity gradients, was utilized to enhance classification performance. Through an ablation study, temperature was identified as a key determinant, with its removal causing a 14.3% accuracy drop for fish and shrimp. Similarly, omitting chlorophyll concentration led to an 11.8% increase in seaweed misclassification, highlighting its role in distinguishing autotrophic from heterotrophic organisms. The proposed ResNet-KAN model demonstrated superior performance compared to conventional deep learning architectures, achieving an overall classification accuracy of 94.6%. Computational efficiency analysis revealed a complexity of $$O\left({N}^{2}L+DL\right)$$ , with processing times scaling predictably across species categories. These findings underscore the importance of integrating remote sensing-derived environmental predictors for robust marine classification. The proposed approach provides a scalable, high-accuracy solution for marine resource monitoring, with potential applications in ecosystem management and precision aquaculture. Additionally, this study includes an exploration of the economic and nutritional implications of the Free Nutritious Meals (MBG) program in Indonesia, assessing alternative strategies for improving its impact on public health and food security.