The detection of construction in agricultural areas or vegetation using satellite images depends on satellite that is used and the type of monitors that use this satellite.
For example,LANDSAT uses monitors with RGB 345, 347 RGB and RGB 234 combinations for environmental sectors and RGB 157, 257 RGB, RGB 145, 247 RGB and RGB 245 for the classification of mulches.
Please, tell me which satellite images you use to detect vegetation or dispersed construction in agricultural areas.
One possible way is to use NDVI (Normalized Difference Vegetation Index). There are many publications/papers that tell you about NDVI (e.g. http://earthobservatory.nasa.gov/Features/MeasuringVegetation/measuring_vegetation_2.php ) and how to calculate NDVI using a satellite images. These publications also tell you about possible values for different land cover types.
I want to identify agricultural land use changes to other land uses, especially scattered buildings or constructions (Small houses and roads) using Landsat satellite images,
Its goog if first you make NDVI and than give tranning sites in supervised (MXL) classification. In classificed image, you will find all small and big settlements.
I would recommend the use of Spectral Unmixing,, then you can identfy the pixels with the bigger amount of substrate fraction in the case of construction, and possibly use the vegetation fraction image to map the percentage amount of vegetation in each pixel.