Satellite image classification plays an important role in surveillance, detection, land planning and management etc. Traditional stacked layered CNN architecture with high learnable parameters faces overfitting issues. Moreover, CNN model having single classifier is unable to make diversity of the prediction which affect the model’s performance. In this paper, an efficient CNN based ensemble model with multiple classifiers is proposed for satellite image classification as multiple classifiers have their own strength and weakness. The proposed ensemble model utilizes Softmax, SVM, KNN, RF, and NB classifiers to improve diversity of model where one classifier correct errors made by others. To overcome overfitting issues, a global average pooling and a dropout layer are incorporated without introducing any learnable parameters in the model. In preprocessing, datasets are resized into 64 × 64 pixels to make equal dimension and sharpened to highlight geographical changes and patterns. The proposed ensemble model is trained on two publicly available dataset such as RSI-CB256 and EuroSAT, showing notable accuracy of 99.11% and 93.52% than model with single classifier, respectively. Additionally, the proposed model presents a lightweight approach with less than 0.31 million parameters and only 14 layers without utilizing traditional stacked layer approach, demonstrating it suitable for satellite image classification.