A significant priority for hyperspectral unmixing is figuring out the ideal number of endmember sources, often known as “virtual dimensionality” (VD). Despite having a direct impact on HU results, VD prediction is frequently handled independently of the HU procedure. To mutually predict the VD throughout the object separation process, the saliency-based autonomous endmember detection technique is introduced in this paper. Pure images have larger availability abnormality (AA) values than “misdirections,” thus we first demonstrate that this is a significant aspect of unnoticed subimages. In order to distinguish identified components from sound, SAED incorporates a hyperpixel earlier. This takes cues from the idea that elim frequently assemble in particular nearby regions within the picture as a result of adjacent pixels. The EE is presented as a strong area prediction task, the unknown fragmented are specifically defined as sensory elements in the AA sector, and the VD is simply figured out if there were no bright items in the AA image space. A recommended process uses the understanding and abilities of the subimages, which makes it more trustworthy than wavelength techniques. Mostly with real and false infrared imaging pictures, this was validated.