Determining and classifying nitrogen deficiency is important for tomato planting. A nitrogen deficiency classification model based on the leaf color features of tomato was proposed. The accuracy of the proposed model can reach over 0.80. The leaf surface of tomatoes planted in summer were covered with glandular hairs. The glandular hairs were conducive to the absorption of water and nutrient elements in a tomato leaf. Under the same concentration of a nutrient solution, the yellowing process of these leaves was different from that of leaves without glandular hairs. Therefore, the accuracy of the classification model based only on leaf color features was reduced to 0.65. The two shape features, namely, the circumference and area of the hair-covered tomato leaves, were both smaller than those of the hairless tomato leaves. Thus, the two shape features of tomato leaf combined with the original leaf color features were used as model inputs to build a new nitrogen deficiency classification model for tomato. The image acquisition unit was constructed using Raspberry Pi and its camera module. Wireless data transmission among smartphones, image acquisition units and local computers was completed using WiFi or a 4G network. Smartphones remotely controlled the acquisition of images and transferred the obtained images through the Web interface to a cloud platform for storage. The local computer preprocessed the images to extract the leaf shape and color features, input the model for prediction, and output the prediction result. The test results showed that the image acquisition system worked properly with temperature ranging from 19.7℃ to 28.3℃ in spring and summer, and the illumination was in the range of 1.125~9.543lx. Preprocessing and segmentation of the acquired images removed any influence of the environment. Using the optimized weighted random forest model, the accuracy of the leaf nitrogen classification model based on shape and color features reached 0.83.