Search

Abstract Most existing multi‐view stereo (MVS) methods fail to consider global context information in the stage of feature extraction and cost aggregation. As transformers have shown remarkable performance on various vision tasks due to their ability to perceive global contextual information, this paper proposes a transformer‐based feature enhancement network (TF‐MVSNet) to facilitate feature representation learning by combining local features (both 2D and 3D) with long‐range contextual information. To reduce memory consumption of feature matching, the cross‐attention mechanism is leveraged to efficiently construct 3D cost volumes under the epipolar constraint. Additionally, a colour‐guided network is designed to refine depth maps at a coarse stage, hence reducing incorrect depth predictions at a fine stage. Extensive experiments were performed on the DTU dataset and Tanks and Temples (T&T) benchmark and results are reported.

Lack of sufficient and balanced data is one of the major challenges in hyperspectral image classification. This problem can cause poor classification performance, especially in detecting or classifying samples of minority classes. The easiest way to overcome the problem is by resampling or creating synthetic samples to balance the class distributions. As the most advanced generative method, generative adversarial networks (GANs) have been used for generating synthetic data. However, GANs need a large amount or sufficient minority class data to train. In this article, we propose to leverage the synthetic minority oversampling technique (SMOTE) in GANs for creating high quality synthetic data to tackle the imbalance problem. The main idea is to train the generator of the GAN to synthesize data from pattern vectors instead of random noise vectors so to guide the GAN to produce data that can expand the minority class data on the decision boundaries. We used kernel principal component analysis and SMOTE to create the pattern vectors and used a silhouette score to control and prevent overlapping issues. In addition, we applied a self-attention module and an automatic data filter to further minimize potentially wrongly labeled or overlapping samples before being added into the training set. Experimental results on both hyperspectral and remote sensing datasets show that the proposed technique can generate more realistic, diverse, and unambiguous synthetic data, resulting in significantly improved classification performances over the existing oversampling techniques.

Abstract Cassava brown streak disease (CBSD) is an emerging viral disease that can greatly reduce cassava productivity, while causing only mild aerial symptoms that develop late in infection. Early detection of CBSD enables better crop management and intervention. Current techniques require laboratory equipment and are labour intensive and often inaccurate. We have developed a handheld active multispectral imaging (A-MSI) device combined with machine learning for early detection of CBSD in real-time. The principal benefits of A-MSI over passive MSI and conventional camera systems are improved spectral signal-to-noise ratio and temporal repeatability. Information fusion techniques further combine spectral and spatial information to reliably identify features that distinguish healthy cassava from plants with CBSD as early as 28 days post inoculation on a susceptible and a tolerant cultivar. Application of the device has the potential to increase farmers’ access to healthy planting materials and reduce losses due to CBSD in Africa. It can also be adapted for sensing other biotic and abiotic stresses in real-world situations where plants are exposed to multiple pest, pathogen and environmental stresses.

Change detection based on multi-temporal synthetic aperture radar (SAR) images plays a significant role in environmental and earth observations. With the advancement in deep neural networks, existing work in the literature mainly concentrates on developing self-supervised methods to generate pseudo-labeled samples to guide the subsequent deep learning based detection. However, this way of selecting sample inevitably introduces erroneous labels and imbalance between unchanged and changed classes, thus causing deterioration in change detection performance. To mitigate these issues, we have proposed a SAR change detection network based on visual saliency and multi-hierarchical fuzzy clustering. Specifically, with multi-dimensional difference feature representations, a visual saliency based difference map is constructed for accurate difference feature extraction. By integrating neighborhood information and hierarchical clustering, the multi-hierarchical fuzzy local information C-means clustering algorithm has been developed to identify potential changed regions for sample selection. A class-balanced adaptive focal loss has further been incorporated into the network training to obtain accurate predictions. Extensive experiments and comparisons on five datasets have been performed. The proposed method has achieved averaged accuracy of 99.07% and Kappa coefficient of 79.87%, outperforming other state-of-the-art algorithms both visually and quantitatively.

Abstract Background The use of spectral imaging within the plant phenotyping and breeding community has been increasing due its utility as a non-invasive diagnostic tool. However, there is a lack of imaging systems targeted specifically at plant science duties, resulting in low precision for canopy-scale measurements. This study trials a prototype multispectral system designed specifically for plant studies and looks at its use as an early detection system for visually asymptomatic disease phases, in this case Pyrenopeziza brassicae in Brassica napus. The analysis takes advantage of machine learning in the form of feature selection and novelty detection to facilitate the classification. An initial study into recording the morphology of the samples is also included to allow for further improvement to the system performance. Results The proposed method was able to detect light leaf spot infection with 92% accuracy when imaging entire oilseed rape plants from above, 12 days after inoculation and 13 days before the appearance of visible symptoms. False colour mapping of spectral vegetation indices was used to quantify disease severity and its distribution within the plant canopy. In addition, the structure of the plant was recorded using photometric stereo, with the output influencing regions used for diagnosis. The shape of the plants was also recorded using photometric stereo, which allowed for reconstruction of the leaf angle and surface texture, although further work is needed to improve the fidelity due to uneven lighting distributions, to allow for reflectance compensation. Conclusions The ability of active multispectral imaging has been demonstrated along with the improvement in time taken to detect light leaf spot at a high accuracy. The importance of capturing structural information is outlined, with its effect on reflectance and thus classification illustrated. The system could be used in plant breeding to enhance the selection of resistant cultivars, with its early and quantitative capability.

Blind image deconvolution/deblurring (BID) is a challenging task due to lack of prior information about the blurring process and image. Noise and ringing artefacts resulted during the restoration process further deter fine restoration of the pristine image. These artefacts mainly arise from using a poorly estimated point spread function (PSF) combined with an ineffective restoration filter. This paper presents a BID scheme based on the steepest descent in kurtosis maximization. Assuming uniform blur, the PSF can be modelled by a parametric form. The scheme tries to estimate the blur parameters by maximizing kurtosis of the deblurred image. The scheme is devised to handle any type of blur that can be framed into a parametric form such as Gaussian, motion and out-of-focus. Gradients for the blur parameters are computed and optimized in the direction of increasing kurtosis value using a steepest descent scheme. The algorithms for several common blurs are derived and the effectiveness has been corroborated through a set of experiments. Validation has also been carried out on various real examples. It is shown that the scheme optimizes on the parameters in a close vicinity of the true parameters. Results of both benchmark and real images are presented. Both full-reference and non-reference image quality measures have been used in quantifying the deblurring performance. The results show that the proposed method offers marked improvements over the existing methods.