Diagnosis of diabetic retinopathy (DR) involves visual examination of retinal images by ophthalmologist to detect pathological signs such as exudate, haemorrhage (HEM) and microaneurysm (MA). This process is conducted manually, therefore it is time-consuming and subjected to human error. This paper develops an automatic and intelligent machine learning algorithm for the detection of diabetic retinopathy (DR) in fundus image. It involves image enhancement and classification of pathological signs using convolution neural network (CNN) for the DR pathological signs classification. In the image enhancement process, high-pass filter and histogram equalization are applied to improve visual quality of fundus images. A five layers CNN architecture is implemented to classify the three pathological signs; exudate, HEM and MA. Two dataset, DIARETDB1 and e-Ophtha are used to evaluate the performance of the system. Simulation results using enhanced DR images show significant improvement in classification accuracy compared to those images without enhancement for both datasets.