Some extent of success can be expected by using data from IRS satellites including latest resocurcesat-2. Since these sensors are multispectral, limited accuracy can be possible.

Use supervise algorithm for producing better results.

The accuracy might be further improved by integrating object based classification with spectral data 

You could use this approach: There exist spectral signatures for different crops at different stages of growth. If you can get spectral signatures of the mango at the stage of growth, you want you could use to identify the mango coverage from the satellite data. 

Your question suggests that you intend to carry out two tasks; (1) classifying horticultural crops and (2) estimating yield. For the first task I suggest you use a combination of Hyperspectral and Multispectral Remote Sensing. You can perform a good classification with any high resolution mu-ltispectral satellite images such as LandSat, SPOT or IKONOS. But to achieve an optimum result you will need to complement your classification with hyper-spectral remote sensing. This entails the use of ground-based instrument such as a spectroradiometer. As for the second task you may use any of the Vegetation indices such as NDVI, LAI, SAVI etc. Good luck! 

Ensemble Classifiers