Seismic image interpretation is indispensable for oil and gas industry. Currently, artificial intelligence has been undertaken to increase the level of confidence in exploratory activities. Detecting potentially recoverable hydrocarbon zones (leads) under the viewpoint of computer vision is an emerging problem that demands thorough examination. This paper introduces a processing workflow to recognize geologic leads in seismic images that resorts to encoder-decoder architectures of a convolutional neural network (CNN) accompanied by segmentation maps and post-processing operations. We have used seismic images collected at offshore sites of the Sergipe-Alagoas Basin (northeast of Brazil) as input. After performing a patch-based data augmentation, a total of 29600 patches were achieved. Out of these, 24000 were used for training, 5000 for validation, and 600 for testing. Each image generated for the training set was post-processed through reconstruction, thresholding-binarization and deblurring-, and outlier removal. By using the dice loss function, intersection-over-union index, and relative areal residual computed after intense cross-validation training rounds, we have shown that the accuracy of the network to detect leads was higher than 80%. Furthermore, the validation error limits were found stable within 5%-10% in all validation rounds, thereby resulting in a fairly accurate prediction of the pre-labelled hydrocarbon spots. INDEX TERMS Encoder-decoder, petroleum exploration, segmentation, seismic imaging. I. INTRODUCTION Spotting hydrocarbon reservoirs accurately is one of the leading problems faced by geologists who have the tricky responsibility to read and interpret seismic images. When seeking some evidence that spurs exploratory activities to achieve successful prospects, much effort and time are spent with the acquisition, analysis, and interpretation of geophysical and geological data [1]. In the petroleum industry, well drilling is costly and strongly dependent on human decisions. Consequently, exhaustive rounds of image interpretation are undertaken [2] to minimize potential failures that eventually are caused by subjective judgments [3]. In a few cases, this may incur in operational downsides. Over the last decades, the traditional seismic interpretation evolved from essential qualitative analyses to consistent quantitative methods, such as post-stack amplitude analysis (e.g. direct hydrocarbon indicators, bright-spots, dim-outs), offset-17 dependent amplitude analysis (AVO analysis), acoustic and 18 elastic impedance inversion, and forward seismic modeling 19 [4]. These methods are strongly supported by maps, such 20 as time structure, depth, and seismic amplitude maps. In 21 particular, self-organizing maps (SOMs) are useful for data 22 clustering and applicable in facies mapping when information 23 about the subsurface is scarce [5]. 24 So far, reflection seismology is the commonplace method 25 used to identify hydrocarbon reservoirs. In this process, the 26 subsurface rocks are represented by dif...
