In view of the inefficient manual cleaning of solar panels and the automatic cleaning robot will cause waste of resources. This paper proposes a solar panels pollution alarm system based on convolution neural network. Collecting contaminated and uncontaminated images of solar panels as training set to train convolution neural network, embedding the model into the website. The website displays the alarm message. The experimental results show that the classification accuracy of the system is 97.6%, the system worls stably and has strong practical value.#$NLKeywords: Photovoltaic system; convolution neural network; alarm system; website