Search

Accurate segmentation of colonoscopic polyps is considered a fundamental step in medical image analysis and surgical interventions. Many recent studies have made improvements based on the encoder-decoder framework, which can effectively segment diverse polyps. Such improvements mainly aim to enhance local features by using global features and applying attention methods. However, relying only on the global information of the final encoder block can result in losing local regional features in the intermediate layer. In addition, determining the edges between benign regions and polyps could be a challenging task. To address the aforementioned issues, we propose a novel separated edge-guidance transformer (SegT) network that aims to build an effective polyp segmentation model. A transformer encoder that learns a more robust representation than existing convolutional neural network-based approaches was specifically applied. To determine the precise segmentation of polyps, we utilize a separated edge-guidance module consisting of separator and edge-guidance blocks. The separator block is a two-stream operator to highlight edges between the background and foreground, whereas the edge-guidance block lies behind both streams to strengthen the understanding of the edge. Lastly, an innovative cascade fusion module was used and fused the refined multi-level features. To evaluate the effectiveness of SegT, we conducted experiments with five challenging public datasets, and the proposed model achieved state-of-the-art performance.

The detection of cracks in sleepers is important in ensuring safe operation of trains and the crack width is a key index in this process. The traditional YOLOv5 model can not achieve satisfactory detection results in the face of large changes in the scale of the target and interference of background information in the detection process. In this paper, a method to detect and measure the fine crack of sleeper based on improved YOLOv5 model of cascade fusion and edge fitting algorithm is proposed to solve these problems. An adaptive multi-scale feature fusion technique is proposed to achieve multi-scale feature fusion efficiently and solve the problem of large changes in the scale of the target. An improved method of feature recombination of attention mechanism is used to solve the interference problem of background information. On this basis, a method based on cascade fusion is proposed to fuse the above improvements and reconstruct the structure of the YOLOv5 model. Moreover, a fitting algorithm of the crack edge is proposed to realize the precise measurement of the maximum width of the fine crack on sleepers. Finally, the process of crack detection and width measurement of sleeper is developed and the performance of the method proposed in this paper is verified by experiments through self-made data sets. The experimental results show that the precision of crack detection is increased from 84.08% to 96.86%, and the recall is increased from 78.13% to 96.46%. The increases are 12.78% and 18.33% respectively. At the same time, the average detection time is reduced from 15.23 ms to 10.04 ms. The measurement accuracy of the maximum crack width is 0.05 mm. [ABSTRACT FROM AUTHOR]

Crowd counting is considered as the essential computer vision application that uses the convolutional neural network to model the crowd density as the regression task. However, the vision-based models are hard to extract the feature under low-quality conditions. As we know, visual and audio are used widely as media platforms for human beings to touch the physical change of the world. The cross-modal information gives us an alternative method of solving the crowd counting task. In this case, in order to solve this problem, a model named the Audio-Visual Multi-Scale Network (AVMSN) is established to model the unconstrained visual and audio sources for completing the crowd counting task in this paper. Based on the Feature extraction and Multi-modal fusion module, in order to handle the objects of various sizes in the crowd scene, the Sample Convolutional Blocks are adopted by the AVMSN as the multi-scale Vision-end branch in the Feature extraction module to calculate the weighted-visual feature. Besides, the audio, which is the temporal domain transformed into the spectrogram information and the audio feature is learned by the audio-VGG network. Finally, the weighted-visual and audio features are fused by the Multi-modal fusion module, which adopts the cascade fusion architecture to calculate the estimated density map. The experimental results show the proposed AVMSN achieves a lower mean absolute error than other state-of-art crowd counting models under the low-quality conditions.

Low-light image enhancement has been an important research branch in the field of computer vision. Low-light images are characterized by poor visibility, high noise and low contrast. To improve low-light images generated in low-light environments and night conditions, we propose an Attention-Guided Multi-scale feature fusion network (MSFFNet) for low-light image enhancement for enhancing the contrast and brightness of low-light images. First, to avoid the high cost computation arising from the stacking of multiple sub-networks, our network uses a single encoder and decoder for multi-scale input and output images. Multi-scale input images can make up for the lack of pixel information and loss of feature map information caused by a single input image. The multi-scale output image can effectively monitor the error loss in the image reconstruction process. Second, the Convolutional Block Attention Module (CBAM) is introduced in the encoder part to effectively suppress the noise and color difference generated during feature extraction and further guide the network to refine the color features. Feature calibration module (FCM) is introduced in the decoder section to enhance the mapping expression between channels. Attention fusion module (AFM) is also added to capture contextual information, which is more conducive to recovering image detail information. Last, the cascade fusion module (CFM) is introduced to effectively combine the feature map information under different perceptual fields. Sufficient qualitative and quantitative experiments have been conducted on a variety of publicly available datasets, and the proposed MSFFNet outperforms other low-light enhancement methods in terms of visual effects and metric scores. [ABSTRACT FROM AUTHOR]

Low-light image enhancement has been an important research branch in the field of computer vision. Low-light images are characterized by poor visibility, high noise and low contrast. To improve low-light images generated in low-light environments and night conditions, we propose an Attention-Guided Multi-scale feature fusion network (MSFFNet) for low-light image enhancement for enhancing the contrast and brightness of low-light images. First, to avoid the high cost computation arising from the stacking of multiple sub-networks, our network uses a single encoder and decoder for multi-scale input and output images. Multi-scale input images can make up for the lack of pixel information and loss of feature map information caused by a single input image. The multi-scale output image can effectively monitor the error loss in the image reconstruction process. Second, the Convolutional Block Attention Module (CBAM) is introduced in the encoder part to effectively suppress the noise and color difference generated during feature extraction and further guide the network to refine the color features. Feature calibration module (FCM) is introduced in the decoder section to enhance the mapping expression between channels. Attention fusion module (AFM) is also added to capture contextual information, which is more conducive to recovering image detail information. Last, the cascade fusion module (CFM) is introduced to effectively combine the feature map information under different perceptual fields. Sufficient qualitative and quantitative experiments have been conducted on a variety of publicly available datasets, and the proposed MSFFNet outperforms other low-light enhancement methods in terms of visual effects and metric scores.

Florescence information monitoring is essential for strengthening orchard management activities, such as flower thinning, fruit protection, and pest control. A lightweight object recognition model using cascade fusion YOLOv4-CF is proposed, which recognizes multi-type objects in their natural environments, such as citrus buds, citrus flowers, and gray mold. The proposed model has an excellent representation capability with an improved cascade fusion network and a multi-scale feature fusion block. Moreover, separable deep convolution blocks were employed to enhance object feature information and reduce model computation. Further, channel shuffling was used to address missing recognition in the dense distribution of object groups. Finally, an embedded sensing system for recognizing citrus flowers was designed by quantitatively applying the proposed YOLOv4-CF model to an FPGA platform. The mAP@.5 of citrus buds, citrus flowers, and gray mold obtained on the server using the proposed YOLOv4-CF model was 95.03%, and the model size of YOLOv4-CF + FPGA was 5.96 MB, which was 74.57% less than the YOLOv4-CF model. The FPGA side had a frame rate of 30 FPS; thus, the embedded sensing system could meet the demands of florescence information in real-time monitoring.

Accurate segmentation of colonoscopic polyps is considered a fundamental step in medical image analysis and surgical interventions. Many recent studies have made improvements based on the encoder-decoder framework, which can effectively segment diverse polyps. Such improvements mainly aim to enhance local features by using global features and applying attention methods. However, relying only on the global information of the final encoder block can result in losing local regional features in the intermediate layer. In addition, determining the edges between benign regions and polyps could be a challenging task. To address the aforementioned issues, we propose a novel separated edge-guidance transformer (SegT) network that aims to build an effective polyp segmentation model. A transformer encoder that learns a more robust representation than existing convolutional neural network-based approaches was specifically applied. To determine the precise segmentation of polyps, we utilize a separated edge-guidance module consisting of separator and edge-guidance blocks. The separator block is a two-stream operator to highlight edges between the background and foreground, whereas the edge-guidance block lies behind both streams to strengthen the understanding of the edge. Lastly, an innovative cascade fusion module was used and fused the refined multi-level features. To evaluate the effectiveness of SegT, we conducted experiments with five challenging public datasets, and the proposed model achieved state-of-the-art performance.

Florescence information monitoring is essential for strengthening orchard management activities, such as flower thinning, fruit protection, and pest control. A lightweight object recognition model using cascade fusion YOLOv4-CF is proposed, which recognizes multi-type objects in their natural environments, such as citrus buds, citrus flowers, and gray mold. The proposed model has an excellent representation capability with an improved cascade fusion network and a multi-scale feature fusion block. Moreover, separable deep convolution blocks were employed to enhance object feature information and reduce model computation. Further, channel shuffling was used to address missing recognition in the dense distribution of object groups. Finally, an embedded sensing system for recognizing citrus flowers was designed by quantitatively applying the proposed YOLOv4-CF model to an FPGA platform. The mAP@.5 of citrus buds, citrus flowers, and gray mold obtained on the server using the proposed YOLOv4-CF model was 95.03%, and the model size of YOLOv4-CF + FPGA was 5.96 MB, which was 74.57% less than the YOLOv4-CF model. The FPGA side had a frame rate of 30 FPS; thus, the embedded sensing system could meet the demands of florescence information in real-time monitoring. [ABSTRACT FROM AUTHOR]