You didnt tell the image u are going to use. If u use Landsat then it is difficult to interpret perennial crops with NDVI. U may classified the image and collect classes. Then u can do stratified sampling in the field. Samples size should be at lest 10% out of the total area of a class distributing covering the area. For all the classes u should apply the same field based method. Finally for accuracy assessment u may use Google earth data and data u have collected. Do Cappa accuracy. details given are not sufficient for any further comments.

For NDVI calculation using UAV or satellite image, you may refer to the attached references. NDVI calculation and its interpretation using satellite image processing might be tricky. While the value ranges between -1 and +1, you will receive different values depending upon the level of pre-processing. It is better to convert the raw DN values to reluctance values before using the NDVI formula. Regarding sample size selection, you may use 10% or more sampling intensity, but it essentially depends on your acceptable margin of error.

Tamasha, I don't want to use an satellite image. I would prefer to use something like a "camera" and so, do my own images