Integrated gravity and gravity gradient 3-D inversion can effectively improve the resolution of inversion results. This study aims to find a better way to extract geological information from gravity and gravity gradient data and to obtain the recovered model with a higher resolution. We present three different methods for integrating gravity and gravity gradient data in inversion: directly integrating gravity and gravity gradient data (DIGG), the sequential inversion of gravity and gravity gradient data (SIGG), and integrating gravity and gravity gradient components using a weighting matrix (IGGW). In addition, we propose a new discrete method for smooth inversion. We discretize the stabilizing functional into two parts. One is the square of the 2-norm of the model parameter, and the other one is the square of the Frobenius norm of the gradient of the model parameter. We adopt a conjugate gradient algorithm (CGA) to solve the objective function. We used the DIGG, SIGG, and IGGW inversion methods in a dip-slab model with multiple anomalous bodies. The inversion results show that by integrating gravity and gravity gradient in inversion, more reasonable results are obtained than by inverting a single data type. Among the presented methods, the IGGW method achieved the best performance in using information contained in gravity and gravity gradient data, followed by the SIGG method, and finally, by the DIGG method. We used the methods on actual data from Vinton salt dome, southwestern Louisiana, USA. The obtained results support the abovementioned conclusions.

Lithology identification is a crucial step in geological research. In recent years, the development of artificial intelligence technologies has provided new insights into solving problems associated with subjectivity and labor intensity of traditional manual identification. However, when rocks are identified in situ, existing algorithms cannot accurately identify them if the image features of different types of rocks are similar or the rock textures are varied. In this regard, the study of lithology identification for the rock images captured from the field was carried out. First, the object detection algorithm of single shot multibox detector was improved by adding residual net and adaptive moment estimation, and a lithology identification model was constructed. Second, based on the above improved algorithm, the technologies of database and geographic information system were combined to develop an integrated identification method. Third, the proposed methods were applied to 12 types of rocks in Xingcheng area, China, for testing their validity, and feasibility in field geological surveys. Finally, the effects of learning rate and batch size on the identification were discussed, as the epoch number was increased. We found that the average accuracies of the improved single shot multibox detector and integrated method were 89.4% and 98.4%, respectively. The maximum accuracy could even reach 100%. The identification results were evaluated based on accuracy, precision, recall, F₁-score, and mean average precision. It was demonstrated that the integrated method has a strong identification ability compared with other neural network methods. Generally, a small learning rate can lead to low loss and high accuracy, whereas a small batch size can lead to high loss and high accuracy. Moreover, the newly proposed methods helped to improve the lithology identification accuracy in the field and support the study of intelligent in situ identification for rock images.

Geophysical joint inversion is more frequently applied to deep crust probes. For the potential field data, it requires the introduction of large-scale observed data and sensitivity matrices. Massive matrix-vector multiplications occur during iterations, and the obtained data cannot be effectively interpreted by merely using desktop computers. To improve the resolution and the computing ability of inversion, we here propose the parallel joint nonlinear inversion of full tensor gravity gradiometry data. As the inversion is affected by linear searches, it is associated with certain classical computing performance issues. Hence, we addressed the memory and efficiency limitations, which are caused by very large calculation volumes. Also, we performed a quantitative and comprehensive feasibility analysis of parallel computing. Then, we identified the main factor influencing the inversion performance and clarified the correspondence between the cell number and the memory. A parallel inversion solution was proposed via graphics processing unit (GPU) based on the sensitivity matrix compression. The data tests demonstrated that inversion has antinoise property and that it can obtain accurate underground density distributions. Also, the parallel solution was found to be suitable for inverting cells at the million cell scale and greater because of its ability of acceleration and matrix compression. A design pattern was applied for gravity or magnetic anomaly inversion of 100 $\times 100\times20$ cells, and the run time was less than 1 min. Overall, we believe that the proposed solution can help implement massive potential field data inversions and promote the application of the parallel technique in other inversion research.

Among single-component inversions, vertical gravity gradient data have been proven to be sufficient to obtain reasonable density distribution. However, the increasing numbers of observations and model parameters inevitably produce a large-scale sensitivity matrix, causing problems such as large memory occupation and slow computations in the inversion. Aiming at solving these long-standing problems, we have introduced a matrix compression technique to reduce memory occupation and constructed a hybrid parallel algorithm for accelerating focusing inversion. We develop the source codes and the design pattern for the proposed algorithm using C language with open multiprocessing and compute unified device architecture for further improving algorithm feasibility and reusability. A graphical user interface and visualization tools also are developed using Python. In the tests with synthetic data produced by numerical simulation and real data from Vinton Dome, it is verified that the inversion can produce an accurate result of density values and object location. The performance analysis is conducted using different scales of data. The speedup increases with the increasing number of graphics processing units (GPUs) and model parameters, proving that the proposed parallel algorithm is scalable. We compare this algorithm with others without matrix compression to demonstrate how the compressed sensitivity matrix storing and accessing method influences inversion performance. More specifically, for an inversion with 100 × 100 × 20 cells, the average speed is approximately 6 s per iteration, with a maximum speedup of 36.94. Our algorithm has a short running time and high multiple-GPU parallel efficiency. Hence, the hybrid parallel algorithm is more powerful at implementing large-scale data calculations and could be used for solving computing problems in mass data inversions.