Search

The increasing interest in filter pruning of convolutional neural networks stems from its inherent ability to effectively compress and accelerate these networks. Currently, filter pruning is mainly divided into two schools: norm-based and relation-based. These methods aim to selectively remove the least important filters according to predefined rules. However, the limitations of these methods lie in the inadequate consideration of filter diversity and the impact of batch normalization (BN) layers on the input of the next layer, which may lead to performance degradation. To address the above limitations of norm-based and similarity-based methods, this study conducts empirical analyses to reveal their drawbacks and subsequently introduces a groundbreaking complex hybrid weighted pruning method. By evaluating the correlations and norms between individual filters, as well as the parameters of the BN layer, our method effectively identifies and prunes the most redundant filters in a robust manner, thereby avoiding significant decreases in network performance. We conducted comprehensive and direct pruning experiments on different depths of ResNet using publicly available image classification datasets, ImageNet and CIFAR-10. The results demonstrate the significant efficacy of our approach. In particular, when applied to the ResNet-50 on the ImageNet dataset, achieves a significant reduction of 53.5% in floating-point operations, with a performance loss of only 0.6%. [ABSTRACT FROM AUTHOR]

The recognition of fish species holds significant importance in aquaculture and marine biology. However, it is a challenging problem due to the high similarity among intra-genus species. Existing recognition methods primarily seek prominent features of the species. However, we believe that the diverse levels of similarity between a species and other species can also function as implicit characteristics for that specific species. Based on this perspective, we propose a dual-branch fusion network for fine-grained fish species recognition utilizing inter-species similarity. This approach consists of a backbone network and two branches for coarse- and fine-grained recognition. In the coarse-grained branch, we designed a guidance matrix and species similarity labels to facilitate the generation of species similarity information. In the fine-grained branch, features from the backbone network are fused with similarity information to achieve precise recognition. Finally, fine-tuning the neural network through loss functions. We conduct experimental validation on three publicly available fish datasets, yielding excellent accuracy outcomes. Code is available at https://github.com/xingxing317/fish_classification. [ABSTRACT FROM AUTHOR]

Detecting marine object is an attractive but challenging task in ocean exploration and conservation. Although the current popular object detection algorithms perform well in general. Because underwater images are affected by color projections and the scale of the object is usually smaller, the detection performance for marine object is not ideal. Therefore, this paper proposes an augmented weighted bidirectional feature pyramid network (AWBiFPN) that reduces the weakening of underwater image features and improves the integration efficiency of multi-scale feature, improving marine object detection performance. Specifically, this paper designs a multi-scale feature pyramid for efficient integration of feature information at all levels by combining weighted bidirectional integration pathway with designed Consistent Supervision module. Using Residual Feature Augmentation module to enhance the extraction of unchanging proportional contextual information, reduce the loss of information at the highest level of the feature map in the pyramid network and provide richer particulate characteristics for small-target detection. To scale the feature map and effectively utilize the feature information of different channels in the same spatial location, this article proposes the AWBiFPN layer, which replaces the depthwise separable convolution in the original BiFPN layer with conventional convolution. Evaluated on underwater image dataset (UTDAC), only 12 epochs are trained to 81.63% of mAP; experimented on MS COCO benchmark, achieving 37.9 AP. Meanwhile, the accuracy of the Aquarium Life dataset (AMLD) and underwater detection dataset under natural light (RUOD) are 81.41% mAP and 83.62% mAP, respectively. The results show that the method performed better not only in marine object detection but also in general class detection. • A novel feature pyramid network is proposed for marine object detection. • Three modules are designed to improve detection accuracy and efficiency. • Evaluation tests were conducted on four datasets. • Detection results of various structural designs are displayed. [ABSTRACT FROM AUTHOR]

Traditional object detection methods cannot effectively identify underwater objects with complex backgrounds, and it is difficult to fully obtain the details of small-scale underwater targets, resulting in poor detection performance. We propose a path-augmented Transformer detection framework to address these limitations to explore the semantic details of small-scale underwater targets in complex environments. On one hand, an embedded local path detection information scheme is devised to facilitate the interaction between high-level and low-level features, thereby enhancing the semantic representation of distinctive features of small-scale underwater targets. Rich dependency relationships are established between the acquired high-level and low-level features within the CSWin-Transformer framework, thus fortifying the semantic representation during the encoding phase. Furthermore, a individualized loss function is employed to optimize and fine-tune features at various hierarchical levels. On the other hand, a detection module with flexible and adaptive point representation that is different from conventional square detection methods is designed. This module covers the underwater target from any direction, and the salient point samples in classification localization and feature selection between points realize the feature selection improves the detection accuracy of underwater objects simultaneously. We designed a new weighted loss function to encourage the network to converge better. Experimental results on open-source underwater and remote sensing images of UTDAC, RUOD, and ADios show that the proposed method outperforms other underwater object detection methods in terms of precision(P), recall(R), comprehensive evaluation index of F1-score and FPS(frames per second). • A novel Path Enhanced Transformer is proposed for underwater object detection. • Added semantic representation of salient features of small-scale underwater Objects. • Designed a detection module with flexible, adaptive point representation. • A new weighted loss function is designed. • Evaluation tests were conducted on three datasets. [ABSTRACT FROM AUTHOR]

The phytohormone cytokinin has various roles in plant development, including meristem maintenance, vascular differentiation, leaf senescence, and regeneration. Prior investigations have revealed that cytokinin acts via a phosphorelay similar to the two-component system by which bacteria sense and respond to external stimuli. The eventual targets of this phosphorelay are type-B ARABIDOPSIS RESPONSE REGULATORS (B-ARRs), containing the conserved N-terminal receiver domain (RD), middle DNA binding domain (DBD), and C-terminal transactivation domain. While it has been established for two decades that the phosphoryl transfer from a specific histidyl residue in ARABIDOPSIS HIS PHOSPHOTRANSFER PROTEINS (AHPs) to an aspartyl residue in the RD of B-ARRs results in a rapid transcriptional response to cytokinin, the underlying molecular basis remains unclear. In this work, we determine the crystal structures of the RD-DBD of ARR1 (ARR1RD-DBD) as well as the ARR1DBD-DNA complex from Arabidopsis. Analyses of the ARR1DBD-DNA complex have revealed the structural basis for sequence-specific recognition of the GAT trinucleotide by ARR1. In particular, comparing the ARR1RD-DBD and ARR1DBD-DNA structures reveals that unphosphorylated ARR1RD-DBD exists in a closed conformation with extensive contacts between the RD and DBD. In vitro and vivo functional assays have further suggested that phosphorylation of the RD weakens its interaction with DBD, subsequently permits the DNA binding capacity of DBD, and promotes the transcriptional activity of ARR1. Our findings thus provide mechanistic insights into phosphorelay activation of gene transcription in response to cytokinin. [ABSTRACT FROM AUTHOR]

Extent of resection (EOR) in glioma contributes to longer survival. The purpose of NCT01479686 was to prove whether intraoperative magnetic resonance imaging (iMRI) increases EOR in glioma surgery and benefit survival. Patients were randomized (1:1) to receive the iMRI (n = 161) or the conventional neuronavigation (n = 160). The primary endpoint was gross total resection (GTR); secondary outcomes reported were progression-free survival (PFS), overall survival (OS), and safety. 188 high-grade gliomas (HGGs) and 133 low-grade gliomas (LGGs) were enrolled. GTR was 83.85% in the iMRI group vs. 50.00% in the control group (P < 0.0001). In 321 patients, the median PFS (mPFS) was 65.12 months in the iMRI group and 61.01 months in the control group (P = 0.0202). For HGGs, mPFS was improved in the iMRI group (19.32 vs. 13.34 months, P = 0.0015), and a trend of superior OS compared with control was observed (29.73 vs. 25.33 months, P = 0.1233). In the predefined eloquent area HGG subgroup, mPFS, and mOS were 20.47 months and 33.58 months in the iMRI vs. 12.21 months and 21.16 months in the control group (P = 0.0098; P = 0.0375, respectively). From the exploratory analyses of HGGs, residual tumor volume (TV) < 1.0 cm3 decreased the risk of survival (mPFS: 18.99 vs. 9.43 months, P = 0.0055; mOS: 29.77 vs. 18.10 months, P = 0.0042). LGGs with preoperative (pre-OP) TV > 43.1 cm3 and postoperative (post-OP) TV > 4.6 cm3 showed worse OS (P= 0.0117) It showed that iMRI significantly increased EOR and indicated survival benefits for HGGs, particularly eloquent HGGs. Residual TV in either HGGs or LGGs is a prognostic factor for survival. • Using iMRI improved the chance of GTR in LGGs and HGGs. • iMRI in cerebral glioma surgery translated into prolonged PFS in HGGs. • Achieving GTR might extend PFS and OS of HGGs, particularly for eloquent HGGs. • Residual tumor volume in either HGGs or LGGs is a prognostic factor for survival. [ABSTRACT FROM AUTHOR]

Oil productions can result in pore pressure drop in the reservoir, generating an increase in effective stress and leading to reservoir compaction. The compaction in the subsurface reservoir translates to the earth's surface, which is manifested as a loss of elevation (land subsidence), causing damages to oil production facilities and surface infrastructures. The Liaohe oilfield, located in Liaohe River Delta (LRD), northeast of China, is one of the most significant subsidence areas in China as a direct consequence of oil extraction from the reservoir. Previous studies carried out in this area assumed the oil production-induced displacement retrieved from Interferometric Synthetic Aperture Radar (InSAR) corresponds only to vertical deformation. In this work, for the first time, we proposed a method to retrieve the full three-dimensional (3D) displacement field over the oilfield. We retrieved the vertical and east-west displacement components by combining the multiple-geometry InSAR line-of-sight (LOS) observations and retrieved the north-south component based on the assumption of a physical relationship between the horizontal and vertical displacement. Two ascending and two descending datasets from Sentinel-1 satellite covering the area were processed by an InSAR time series analysis over the 2017 to 2021 period, providing consistent displacement rate maps and displacement time series in the LOS direction. Spatial local-scale land subsidence was found in several producing fields over the deltaic region, including Shuguang, Huanxiling, and Jinzhou oilfields. The 3D displacement decomposition was then conducted in Shuguang oilfield. The derived 3D displacement field exhibit a circular subsidence bowl with a maximum subsiding rate reaching 212 mm/year, accompanied by a centripetal pattern of horizontal displacements with maximum rates up to 50–60 mm/year moving towards the subsidence center. The retrieved 3D displacements are in good agreement with predictions from the geomechanical modeling by assuming a disk-shaped reservoir subject to a uniform reduction in pore fluid pressure. Finally, we show the importance of knowing both the vertical and horizontal displacement in characterizing the lateral boundary of the subsurface reservoir. [ABSTRACT FROM AUTHOR]