Solar panels (photovoltaic panels) are used in various industries, mainly to generate clean electricity and provide energy for various occasions. However, due to long-term exposure to the natural environment, the accumulation of dust on solar panels is inevitable due to factors such as climate, wind, vegetation, and animals, which weakens the ability of solar panels to absorb too much energy, reduces efficiency, shortens lifespan, and increases costs. This paper utilizes a model that combines feature extraction from a convolutional neural network and a machine learning classifier to train and classify the Kaggle solar panel public dataset using the two-class image classification method. The final classification accuracy is 94%, and a system is designed to achieve automatic recognition of solar panel dust detection. This paper adopts a method of combining MobileNet feature extraction with different classifiers, and ultimately, the MobileNet plus logistic regression method has the best effect. The need for regular maintenance of solar panels is high, and the cost is high. This study can help detect dust accumulation on solar panels, reduce costs, improve solar energy utilization, and bring great convenience to solar panel testing.