Image classification plays a significant role in robotic vision. This paper proposes an image classification model: Xception-LightGBM, which combines with Xception and Light Gradient Boosting Machine for hybrid image classification. The proposed algorithm produces the image feature extraction via Xception and classifies these feature vectors using Light Gradient Boosting Machine (LightGBM). The Xception-LightGBM model is compared with five representative image prediction models, such as VGG16, VGG19, InceptionV3, DenseNet121, and Xception. The experiments on six data sets demonstrate this proposed model leads to successful runs and provides optimal performances. It shows this model achieves the best results for all six evaluation metrics: accuracy, precision, recall, F1-Score, loss, and Jaccard. Furthermore, this proposed model acquires the highest accuracy on six image data sets, which has at least 1.1% in accuracy improved to the Xception architecture. It suggests this model may be preferable for robotic vision.