JINGHU JING

Dr. JINGHU JING

orcid

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
27 February 2018 in IEEE Signal Processing Letters
Verified
Journal Article
Closed access
JJJINGHU JING
JINGHU JING
ZCZhibo Chen
Zhibo Chen
MYMeng Yang
Meng Yang
RZRongguo Zhang
Rongguo Zhang
YCYaji Cui
Yaji Cui
Plant leaf recognition is a computer vision task used to automatically recognize plant species. It is very challenging since rich plant leaf morphological variations, such as sizes, textures, shapes, venation, and so on. Most existing plant leaf methods typically normalize all plant leaf images to the same size and recognize them at one scale, resulting in unsatisfactory performances. In this letter, a multiscale fusion convolutional neural network (MSF-CNN) is proposed for plant leaf recognition at multiple scales. First, an input image is down-sampled into multiples low resolution images with a list of bilinear interpolation operations. Then, these input images with different scales are step-by-step fed into the MSF-CNN architecture to learn discriminative features at different depths. At this stage, the feature fusion between two different scales is realized by a concatenation operation, which concatenates feature maps learned on different scale images from a channel view. Along with the depth of the MSF-CNN, multiscale images are progressively handled and the corresponding features are fused. Third, the last layer of the MSF-CNN aggregates all discriminative information to obtain the final feature for predicting the plant species of the input image. Experiments show the proposed MSF-CNN method is superior to multiple state-of-the art plant leaf recognition methods on the MalayaKew Leaf dataset and the LeafSnap Plant Leaf dataset.

15 February 2022 in IET Image Processing
Verified
Journal Article
Open access
PDF file
XLXiaoming Li
Xiaoming Li
HZHengzhi Zhang
Hengzhi Zhang
LFLei Feng
Lei Feng
JJJINGHU JING
JINGHU JING
RZRongguo Zhang
Rongguo Zhang
QQQiang Qiao
Qiang Qiao
Image outpainting aims at extending the field of view of an existing image. While image inpainting has achieved great success with the deep learning technology, image outpainting still receive less attention. The main challenge is how to generate high‐quality extended images with clear texture and highly consistent semantic information. In order to solve the problem of the effect of invalid pixels on the generated image and the distance of effective pixels is too far. This paper proposes a two‐stage image outpainting method (the EA method), which consists of an edge generation stage and an edge transformation stage. In this paper, the convolutional block attention module (CBAM) is introduced into the generation network to focus on spatial and channel feature and the improved VAE‐GAN structure is used to generate the extended image for more realistic semantics. The EA method is evaluated on the CelebA, Pairs‐streetview and homemade landscapes dataset, and show that the results contain high‐quality textures as well as faithfully extend the semantics. The average PSNR, SSIM, FID index of the EA method on the three datasets is 22.7961, 0.7061, 6.8553 and show that it outperforms existing algorithms in both quantitative and qualitative analysis.

1 November 2024 in Engineering Applications of Artificial Intelligence
Verified
Journal Article
Closed access
JCJunhui Cao
Junhui Cao
JJJINGHU JING
JINGHU JING
RZRongguo Zhang
Rongguo Zhang

13 April 2020 in IET Image Processing
Verified
Journal Article
Open access
PDF file
JJJINGHU JING
JINGHU JING
ZCZhibo Chen
Zhibo Chen
RZRongguo Zhang
Rongguo Zhang
MYMeng Yang
Meng Yang
SZShuai Zhang
Shuai Zhang
In this study, the authors focus on the task of leaf segmentation under different imaging conditions (e.g. backgrounds and shadows). A new method ‐ robust random walk (RW) is proposed to propagate the prior of user's specified pixels. Specifically, they first employ RWs to take the relationship of pairwise pixels into consideration. A superpixel‐consistent constraint is added to make the edges of segmentation smooth. Owing to the effect of illumination, some parts of a leaf surface are brighter than others and it may further harm the subsequent label propagation. To address this problem, they learn a common subspace by taking into account the illumination of local and non‐local pixels. By doing so, it has good adaptability to process noise interfering and non‐uniform illumination. In addition, since RW only considers the pairwise relationship of pixels, it will be sensitive to the specified and connected pixels. Thus, they further employ a log‐likelihood ratio to predict the probability of a pixel belonging to the background and use it to guide the label propagation. Based on the proposed method, they can obtain a smoothed and robust leaf segmentation. Experimental results on unconstrained leaf images demonstrate the efficiency of their algorithm.

21 February 2024 in IEEE Access
Verified
Journal Article
Open access
JJJINGHU JING
JINGHU JING
WMWang Meng-Yao
Wang Meng-Yao
WFWang Fei
Wang Fei
ZRZhang Ru-Min
Zhang Ru-Min
LBLian Bing-Quan
Lian Bing-Quan
There are subtle differences between single regions of the same subcategory in fine-grained images. At present, many fine-grained image classification networks often focus on a single region to determine the target category. However, in many cases, most discriminative features in fine-grained images are distributed in multiple local regions of the image, and it is not often enough for fine-grained image to rely solely on one region.To solve these problems, a new method is proposed.This method generates discriminative features through reinforcement learning and obtains complementary regions through complementary network. The reinforcement network and the complementary network learn through adversarial learning and improve the accuracy of fine-grained images classification.The method is tested on CUB200-2011,fine-grained Visual Classification of Aircraft, and Stanford dogs datasets and the results show adequate performance.

17 July 2024 in Research Square
Verified
Preprint
Open access
PDF file
JJJINGHU JING
JINGHU JING
RZRumin Zhang
Rumin Zhang
BLBingquan Lian
Bingquan Lian
FWFang Wang
Fang Wang
Existing deep learning image recognition models heavily rely on large amounts of manually labeled data by professionals, which is difficult to acquire, costly in terms of human labeling, and prone to subjective influence. Additionally, datasets in practical scenarios often exhibit imbalance, with a significant skew towards majority classes over minority classes. Compared to the majority classes, the rare minority classes are often more meaningful. However, most existing models tend to bias towards the majority classes during fitting, resulting in insufficient recognition accuracy for minority classes, which significantly hinders the widespread application of deep learning in practical image recognition. This paper introduces a new semi-supervised generative adversarial network, BASA-GAN, which can realize fair training of each class in unbalanced data set and improve the attention and recognition rate of minority classes. Firstly, we introduce a balanced generator module (SAG) that utilizes an autoencoder to learn features from all classes in the original labeled dataset and generates a balanced dataset through reconstruction. The reconstructed dataset is then fed into the generator to achieve fair ing for each class during training. In SAG we design a completely new class-balanced loss that encourages the autoencoder to generate a more balanced hidden space representation for different classes of data. Then, by incorporating the self-attention mechanism into the SAG module and the discriminator, we enhance the quality of the synthesized images and the discriminative power of the discriminator. Additionally, we introduce the spectral normalization into the semi-supervised model of BASA-GAN to address the instability issues during GAN training. Experimental results demonstrate that the SAG module achieves highly competitive performance in semi-supervised imbalanced image recognition tasks, not only improving the recognition accuracy for minority classes but also achieving fair training for each category in the dataset.