32Gb is not enough to hold your image so it is not possible for you to load the image entirely into your computer's memory.
With each pixel taking up 64 bits (if it is single channel, or multiply by d for d dimensional hyperspectral image) you require approximately 162000 * 105000 * 64 (bits) / (1024^3 bits) = 1013 Gbits (or roughly 126Gb) memory for your image.
You are better off using a GIS image processing library that can perform your image processing tasks out-of-core or only loading in specific subregions of the image. Geoprocessing is a very specific field within image processing and it is best that you use the appropriate libraries for the job. Check this post for more details on using the gdal library for geoprocessing.