Hyperspectral image classification has numerous applications, including land cover mapping, vegetation classification, urban monitoring, and crop mapping. Despite its potential applications, a significant bottleneck for HSI classification remains the insufficient number of training samples. Collecting such labeled training samples is arduous and protracted. To address this issue and motivated by the idea of borrowing and reusing labeled samples from different multiple preexisting deep learning models, this paper introduces a Transfer-Ensemble Learning based Deep Convolutional Neural Network for hyperspectral remote sensing image classification. This study explores dimensionality reduction using Principal Component Analysis and evaluates the performance of six methods: 2D CNN, Mobile Net V2, ResNet 50, VGG16, 3D CNN, and the proposed TEL-DCNN method. All the methods are tested on four standard datasets know as University of Pavia, Indian Pines, Salinas, and Kennedy Space Center. The evaluation metrics used are overall accuracy, Kappa coefficient and average accuracy. The proposed TEL-DCNN method demonstrates promising results, highlighting its potential for advancing HSI classification.