Table 2 The pseudo-algorithm of RapeNet

Input: A UAV-RGB image of one plot in a field.

Output: The number of rape flower clusters in the image and a heat map.

Phase 1: The input image \(K_{i}\) is adjusted to a resolution of 512\(\times\)1024 image \(K_{r}\), set the sliding window resolution 256\(\times\)256, and get 8 sub-images.

Phase 2: The whole backbone network is built by pyramidal convolution. A Bayesian loss function is used to constrain the entire training process. A likelihood function is constructed for each annotated point using a Gaussian.

Phase 3: Train and test the RapeNet network model.