The Sichuan basin holds large potash reserves, making the exploration and evaluation of these resources crucial. Potash exists in two subsurface forms: potassium brine and new polyhalite. Potassium brine is a liquid, high- quality reservoir that is easy to mine. while new polyhalite is a solid form. Due to the lack of a direct relationship between potash reservoirs and seismic data, establishing definitive characteristic equations cannot be built. Therefore, current potash reservoir interpretation relies primarily on analyzing seismic reflection characteristics, lacking specialized prediction technologies. To address this limitation, we propose a convolutional neural network (CNN) incorporating a projection layer, using geological knowledge as prior information. This projected CNN is utilized to obtain characteristic parameters for potash reservoirs. By analyzing potash content and resistivity logs derived from the CNN, we observe distinct characteristics: potassium brine layers exhibit high potash content and low resistivity, while new polyhalite layers display high potash content and high resistivity. Therefore, by comparing predicted potash content and resistivity data from the projected CNN, we can effectively delineate the spatial locations of potassium brine and new polyhalite reservoirs.