1. Convolution can be applied not only on a two-dimensional plane, but also in a high-dimensional space. At this time, the convolution kernel also increases the corresponding dimensions (channels).
The channels of the image and the channels of the convolution kernel must match
2. The convolution kernel is convolved with all the corresponding pixels at the same time
Note: The channel of the convolution result is 1, there are several convolution kernels, and the channel of the result is a few
3. It is also possible to convolve only one of R, G, and B (the rest of the convolution kernel positions are set to 0)
4. If you want to perform edge detection on multiple angles at the same time, use multiple convolution kernels at the same time
Using k convolution kernels at the same time, the resulting channel is k
5. Sometimes channel also becomes depth