Artificial Intelligence-Based Deep Fusion Model for Pan-Sharpening of Remote Sensing Images

During the past two decades, many remote sensing image fusion techniques have been designed to improve the spatial resolution of the low-spatial-resolution multispectral bands. The main objective is fuse the low-resolution multispectral (MS) image and the high-spatial-resolution panchromatic (PAN) image to obtain a fused image having high spatial and spectral information. Recently, many artificial intelligence-based deep learning models have been designed to fuse the remote sensing images. But these models do not consider the inherent image distribution difference between MS and PAN images. Therefore, the obtained fused images may suffer from gradient and color distortion problems. To overcome these problems, in this paper, an efficient artificial intelligence-based deep transfer learning model is proposed. Inception-ResNet-v2 model is improved by using a color-aware perceptual loss (CPL). The obtained fused images are further improved by using gradient channel prior as a postprocessing step. Gradient channel prior is used to preserve the color and gradient information. Extensive experiments are carried out by considering the benchmark datasets. Performance analysis shows that the proposed model can efficiently preserve color and gradient information in the fused remote sensing images than the existing models.

Spatial agents for geological surface modelling

Increased availability and use of 3D-rendered geological models have provided society with predictive capabilities, supporting natural resource assessments, hazard awareness, and infrastructure development. The Geological Survey of Canada, along with other such institutions, has been trying to standardize and operationalize this modelling practice. Knowing what is in the subsurface, however, is not an easy exercise, especially when it is difficult or impossible to sample at greater depths. Existing approaches for creating 3D geological models involve developing surface components that represent spatial geological features, horizons, faults, and folds, and then assembling them into a framework model as context for downstream property modelling applications (e.g. geophysical inversions, thermo-mechanical simulations, and fracture density models). The current challenge is to develop geologically reasonable starting framework models from regions with sparser data when we have more complicated geology. This study explores the problem of geological data sparsity and presents a new approach that may be useful to open up the logjam in modelling the more challenging terrains using an agent-based approach. Semi-autonomous software entities called spatial agents can be programmed to perform spatial and property interrogation functions, estimations and construction operations for simple graphical objects, that may be usable in building 3D geological surfaces. These surfaces form the building blocks from which full geological and topological models are built and may be useful in sparse-data environments, where ancillary or a priori information is available. Critical in developing natural domain models is the use of gradient information. Increasing the density of spatial gradient information (fabric dips, fold plunges, and local or regional trends) from geologic feature orientations (planar and linear) is the key to more accurate geologic modelling and is core to the functions of spatial agents presented herein. This study, for the first time, examines the potential use of spatial agents to increase gradient constraints in the context of the Loop project (https://loop3d.github.io/, last access: 1 October 2021​​​​​​​) in which new complementary methods are being developed for modelling complex geology for regional applications. The spatial agent codes presented may act to densify and supplement gradient as well as on-contact control points used in LoopStructural (https://www.github.com/Loop3d/LoopStructural, last access: 1 October 2021) and Map2Loop (https://doi.org/10.5281/zenodo.4288476, de Rose et al., 2020). Spatial agents are used to represent common geological data constraints, such as interface locations and gradient geometry, and simple but topologically consistent triangulated meshes. Spatial agents can potentially be used to develop surfaces that conform to reasonable geological patterns of interest, provided that they are embedded with behaviours that are reflective of the knowledge of their geological environment. Initially, this would involve detecting simple geological constraints: locations, trajectories, and trends of geological interfaces. Local and global eigenvectors enable spatial continuity estimates, which can reflect geological trends, with rotational bias, using a quaternion implementation. Spatial interpolation of structural geology orientation data with spatial agents employs a range of simple nearest-neighbour to inverse-distance-weighted (IDW) and quaternion-based spherical linear rotation interpolation (SLERP) schemes. This simulation environment implemented in NetLogo 3D is potentially useful for complex-geology–sparse-data environments where extension, projection, and propagation functions are needed to create more realistic geological forms.

Optimization in Milling of Polymer Materials for High Quality Surfaces

High quality surfaces with transparency are required for manufacturing of plastic products. In cutting of polymer materials, surface quality is sometimes deteriorated by tarnish and/or unequal spaces of area on a surface. The cutting parameters should be determined through understanding of surface finish characteristics. This paper presents an optimization approach in milling of polycarbonate with polycrystal diamond tools in terms of the surface finish. Surfaces are finished with changing the feed rate and the clearance angle of the tool. The surface finishes, then, were observed to classify the deterioration type into welding, adhesion, and the unequal space of cutter marks with measurement of the surface profiles. The measured surface roughnesses are decomposed into the theoretical/geometrical term and the irregular term induced by the thermal and the dynamic effects. A map is presented to characterize the irregular term for the feed rates and the clearance angles. Because the surface roughnesses are measured at discrete sets of the cutting parameters in the actual cutting tests, the process design cannot be conducted to optimize the operation parameters. Therefore, a neural network is applied to associate the cutting parameters with the irregular term in the map. An approach is presented to determine the number of hidden nodes/units in the design of the neural network. Three prominent areas of welding, adhesion, and unequal spaces of the cutter marks, appear in the map of irregular roughness. The map of the surface roughness is made to optimize the cutting process. The applicable feed rates and clearance angles are determined for the tolerable surface roughnesses. The gradient information in the map is used to evaluate the stability/robustness of the surface quality for changing the parameters. The optimum parameters were determined to minimize the gradient information in the applicable feed rates and clearance angles.

PP-PG: Combining Parameter Perturbation with Policy Gradient Methods for Effective and Efficient Explorations in Deep Reinforcement Learning

Efficient and stable exploration remains a key challenge for deep reinforcement learning (DRL) operating in high-dimensional action and state spaces. Recently, a more promising approach by combining the exploration in the action space with the exploration in the parameters space has been proposed to get the best of both methods. In this article, we propose a new iterative and close-loop framework by combining the evolutionary algorithm (EA), which does explorations in a gradient-free manner directly in the parameters space with an actor-critic, and the deep deterministic policy gradient (DDPG) reinforcement learning algorithm, which does explorations in a gradient-based manner in the action space to make these two methods cooperate in a more balanced and efficient way. In our framework, the policies represented by the EA population (the parametric perturbation part) can evolve in a guided manner by utilizing the gradient information provided by the DDPG and the policy gradient part (DDPG) is used only as a fine-tuning tool for the best individual in the EA population to improve the sample efficiency. In particular, we propose a criterion to determine the training steps required for the DDPG to ensure that useful gradient information can be generated from the EA generated samples and the DDPG and EA part can work together in a more balanced way during each generation. Furthermore, within the DDPG part, our algorithm can flexibly switch between fine-tuning the same previous RL-Actor and fine-tuning a new one generated by the EA according to different situations to further improve the efficiency. Experiments on a range of challenging continuous control benchmarks demonstrate that our algorithm outperforms related works and offers a satisfactory trade-off between stability and sample efficiency.