Mapping of real Gossan in oxidant deposits using multi-source images and deep learning

Gossans are the easiest and fastest way to explore subsurface resources and actually represent mineral zones on the earth's surface. Gossans that have important mineral resources such as copper and gold are called true gusans. The aim of this study was to identify true Gossans in small exploration areas. In this paper, an algorithm for deep convolutional cane crusts was designed. In the proposed algorithm, first preprocessions such as geometric and spectral correction and restoration, division of satellite images into smaller images and amplification of training data are performed to prepare RGB data to enter the chip. The proposed CNN cane has a encoder-decoder structure that in the coding stage different and efficient features are extracted at different scales and in the decoding stage the generated features are combined to estimate the Gossan regions. Then, the desired network was implemented for the images of the studied exploratory area called "Tal Bargah" located in Darab city and the Gossan areas of the region were extracted. For field evaluation of the obtained results, the results of the network and its location on the copper orthodontic interpolation map of the region and review of the integrated lithological results and the real gusans of the region with statistical accuracy of sensitivity parameters: 0.957, F1 score: 0.457, rock detection accuracy 92% and average Copper grade above 4% was detected in these areas.