Glands are objects of interest which can be used for quantitatively analysis of histology images. Detecting glands from H&E staining histological images based on neural networks, may suffer stain variation problem. In this paper, we present a new method which combines a statistical colour detection model and a neural network to cover this problem. Colours shown at glands boundaries are pre-detected and enhanced in a pre-processing step. Then a neural network model based on Faster R-CNN is learned from these colour pixels to detect glands. This method has been tested on a Colon Histology Images Challenge Contest (GlaS) held at MICCAI 2015. The experimental results have shown the proposed method is superior to either Faster R-CNN or U-net without colour detection pre-processing. In addition, this proposed method can achieve Fl-score rank 8 in detecting benign glands and rank 5 in detecting malignant glands.