The answer is not simple as seems. Using threshold to identify crops is a common application of NDVI, however, its success depend on the spectra difference between the crops or crop state. For example, rice and maize have the same NDVI at specific growing state. NDVI only tell you if the vegetation is vigorous and dense, you need to related these values with complementary data (field data) to determinate which crop is (date, location, etc). On the other hand, for multitemporal application of NDVI, the images have to be "standardized", this can be achieve with an atmospheric correction algorithm (e.g. DOS ). You can also find an appropriate vegetation index at https://www.indexdatabase.de/.
Thank you for reply, definitely local knowledge of crop sowing and harvesting period would also help. Combination of ET with NDVI would also play a role to distinguish crops. indexdatabase would be very helpful indeed.
You can use NDVI from remote sensing images with different temporal resolution such as Landsat 7 and 8 and Sentinel-2 to create a stack of sorted NDVI layers based on the date the acquired images. Based on this stack of NDVI layers you can extract profiles representing the phenological progress of crops (in your agricultural area) to help in distinguishing crops and as spectral signature location from original images.
You can check my papers about this subject on my ResearchGate page.