As one of the important fields of computer vision research, pedestrian attribute recognition has gained increasing attention from domestic and foreign researchers due to its huge potential applications. However, obtaining long-distance pedestrian information in actual scenes poses challenges such as lack of information, incomplete feature extraction, and low attribute recognition accuracy. To address these issues, we propose a multi-scale feature fusion network based on a dual self-attention mechanism. The fusion module merges multi-scale features to enable more complete attribute extraction, while the dual self-attention module focuses the network on important regions. Experimental results on PA-100K, RAP, and PETA datasets achieved mean accuracies of 81.97

Currently, with the widespread use of UAVs in various fields, there is a significant challenge in detecting aerial targets using drones. The YOLO series of object detection has made significant advancements in both speed and accuracy. However, many state-of-the-art methods are not suitable for drone images due to the unique perspective and large number of small targets. Increasing the detection layer or input picture size can improve accuracy, but this also increases computational cost and reduces detection speed. This paper proposes YOLOQ for aerial object detection, utilizing new S-FPN, SPPF+, and S-SIoU modules. Extensive experiments demonstrate that YOLOQ achieves advanced performance on generic drone dataset with only 5.1MB weight files, two million parameters, 13.5GFLOPs of computational complexity, and 192FPS speed.

Pedestrian Attribute Recognition has attracted increasing attention due to its wide range of potential applications. However, the pedestrian images are taken from a far distances significantly increase the difficulty of attribute feature extraction integrity. To solve these problems, and further improve the accuracy of pedestrian attribute recognition by convolutional neural networks, we proposed a method based on multi-scale feature fusion and cross attention. First, we use the ResNeXt50 structure to extract shallow and deep features, then fusion them for improving the recognition accuracy of attributes. Second, we adopt cross attention mechanism to generate pixel-level contextual information for enhancing the important feature information about attributes. Finally, we transmit the feature map into the fully connected layer for classification. Extensive experiments show that our proposed method could achieve state-of-the-art results on three pedestrian attribute datasets, including PA-100K, RAP, and PETA. The corresponding mean accuracies achieve 81.57%, 81.23% and 85.85%,respectively.

As one of the important fields of computer vision research, pedestrian attribute recognition has received increasing attention on researchers at domestic and foreign. However, obtaining long-distance pedestrian information on actual scenes has problems, such as lack of information, incomplete feature extraction, and low attribute recognition accuracy. To address these issues, we proposed a Dual Adaptive Fusion Attention and Criss-Cross Attention Module (DAFCC). This module contains two sub-modules: First, the dual adaptive fusion attention module automatically adjusts the weights of attributes in different scales, then fusion the different scale features and makes attribute extraction more complete. Second, we employ criss-cross attention to extract rich contextual information, which is beneficial for visual understanding. By training on the public PA-100K, RAP and PETA datasets, the mean accuracies achieved 81.09%, 81.44% and 85.94%, respectively. Extensive experimental results show that the method has strong competitiveness among many current classical algorithms.

Current deep forest are mostly built upon multi-grained cascade forest, i.e. a novel decision-tree ensemble, with a cascade structure that enables representation learning by forest. In this paper, we propose the Deep Stacked Random Forest(DSRF) based on Deep Forest. We have made the following three enhancements: (1) The enhancement of data learning strengthens the cascade part of the deep forest model. DSRF applies a Random Feature Extraction (RFE) method to obtain more diverse feature subsets from the original data, which benefits feature learning, thus laying a solid foundation for the performance improvement of the model. (2) Designing the Mean Stacked Forest(MSF), which adopts mean output, reduces the impact of part of the classifiers reaching local optima on the accuracy of the algorithm, reduces the convergence speed of the representation learning model. To a certain extent, MSF reduces memory consumption. (3) Adding the null space matrix on training data promotes the model’s ability to learn mapping rules from the input data, thereby improving the generalization and fault tolerance of the model. Experiment results show that DSRF achieves advanced performance on four low-dimensional datasets and two high-dimensional datasets. Especially, the model achieves state-of-the-art performance in terms of convergence speed and learning efficiency on low-dimensional datasets.

Pedestrian attribute recognition is a vital task in computer vision, which is widely used in pedestrian detection and pedestrian re-identification, etc. Pedestrian attribute recognition aims to excavate the attributes of the target person from video or images. To solve the specific challenging factors in this task, such as changes in viewpoint, lacking illumination, and low resolution, we propose a brand new pedestrian attribute recognition method. Using ASPP to extend the receptive fields and densely connect the generated features, multi-scale feature fusion merges features from different receptive fields into one feature that is more discriminative than the input features. And the AIIM module is used to establish remote dependencies at different spatial scales. Extensive experiments show that our proposed method achieves state-of-the-art results with average accuracy (mA) of 86.35%, 81.60%, and 83.29% on public datasets such as PETA, PA100k, and RAP, respectively.

Quantitatively deriving plant phenotypes from plant images in a non-contact manner is a very challenging task that relies heavily on the accurate segmentation of plant images. Previous methods mainly used the U-Net network structure and attention mechanism to obtain the corresponding plant phenotype segmentation results. However, the U-Net structure and attention mechanism are relatively outdated, and its method can only achieve a Dice score of 98.47% on the open source dataset, which is still insufficient for the recent plant phenotype segmentation task and needs to be further improved for more detailed research. Therefore, in view of the low segmentation performance of existing plant phenotype semantic segmentation models, this paper proposes a semantic segmentation network RT-Net based on an advanced deep learning framework. Specifically, the network mainly adopts the encoder-decoder network structure of deeplabv3+, and the encoding part of the network adopts the more efficient RepVGG as the backbone network for local feature extraction. At the same time, compared with the traditional Atrous Spatial Pyramid Pooling (ASPP), this paper designs the (Atrous Spatial Pyramid Pooling Based Transformer)ASPPBT module to extract more global feature information through a global adaptive method to obtain denser plant phenotypes. The decoding part performs feature fusion on the output of the encoding part, and then uses upsampling to restore the scale, and finally obtains the semantic segmentation result. The experimental results show that the proposed network has achieved a Dice score of 99.33% on the Arabidopsis plant dataset released by the CVPPP14 competition, and has better segmentation ability compared with other advanced plant field segmentation algorithms