This paper takes into account the problem of conversion of RGB images to hyperspectral images, i.e the recovering the entire spectral information/signatures from a 3 channel RGB image. The dataset used here comprises of ‘clean’ images, as in images that are noiseless and are in the PNG format. The dataset consists of 450 clean images along with corresponding 450 hyperspectral images. We took three models, first is Conv_5, second is Enhanced-ResNet and third is Dense-HSCNN. The models get complex in the respective order. In the experimentation results, Conv_5 performed aver-age, the best results were from the Enhanced-ResNet, although due to the lack of computational resources, we couldn’t get the desired results for the Dense-HSCNN model. Having more computational resources and training on more images can reduce the loss in all the three models, especially the Dense-hscnn model, leading us to find more data from the RGB images that can be found in Hyperspectral Images.