In real-life garbage collection tasks, garbage is often distributed in non-uniform patches. To more effectively deal with this patchy garbage, a self-organized collection algorithm for swarm robots based on density interactions is proposed. Firstly, a spatial density field is established to characterize the spatial distribution of garbage and other robots based on the Gaussian kernel function. Secondly, the density field is used to coordinate the collective motion of swarm robots to perform an edge-surrounding behavior in order to encircle the garbage patchy. Finally, in conjunction with the radial shrinkage strategy for coarse-grained garbage and the vortex shrinkage strategy for fine-grained garbage, the scattered garbage is pushed inward in a group collaborative manner by swarm robots, and the collection task of patchy garbage is completed. Numerical simulations and experiments with real swarm robots demonstrate that the proposed garbage collection algorithm is effective for collecting single-patchy garbage, multi-patchy garbage, and dynamic-patchy garbage, showing excellent parallelism and adaptivity.