Traditional data analysis is not up to pace with the quick rate at which multimedia repositories are growing. For finding pertinent images with optimized accuracy, a sophisticated visual informative retrieval (VIR) model is necessary. Currently, the techniques of deep learning are playing a vital role for facing this challenging issue. In addition to ignoring the emphasis on particular channels or locations, deep learning models prioritize the most useful sections of the feature maps. This will result in optimal feature representations of visual images that are less effective since they have various levels of relevance across different channels or regions. This work aimed to address this issue by focusing on improving the feature maps through the use of channels. The experiments conducted with benchmarked datasets for validating the proposed work. The comparative analysis demonstrated that proposed ensemble deep learning technique outperformed the existing deep learning techniques with notable optimized informative query assessment results.