It is a relatively economical, flexible and time-effective method to build a visible light sensor for monitoring of tea growth and diseases, but the resolution of the image will be affected by the flying height of the UAV. Therefore, an improved residual dense network (RDN) for super-resolution reconstruction of UAV tea images was proposed. Specifically, in view of the complex texture of UAV tea images, taking RDN as the baseline network, residual group (RG) was introduced into its structure, combining multiple residual channel attention modules were combined together to treat different channels differently by introducing an attention mechanism, and paying attention to the high-frequency detail information of UAV tea images, thereby improving the representation ability of the network; at the same time, a convolutional long jump structure was designed, using the longrange skip connection with convolution, to dynamically adjust the weight of the feature after passing through the residual dense block (RDB), and making better use of the hierarchical feature information of the UAV tea image, thereby improving the super-resolution of the quality of reconstructed image. The experimental results showed that the improved RDN network performed better than other algorithms on the test set of UAV tea images, and the super-resolution reconstructed images had higher peak signal-to-noise ratio and structural similarity. In the case of quadruple super resolution, it can reach 36.03dB and 0.9132, respectively, which can provide support for the followup research of tea intelligent monitoring.