Xiang Fan

Xiang Fan


Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

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

Representative Publications

View All Publications
5 January 2025 in Sensors
Verified
Journal Article
Open access
PDF file
舒刘舒宽 刘
舒宽 刘
SCShi Cao
Shi Cao
XLXia Lu
Xia Lu
JPJiqing Peng
Jiqing Peng
LPLina Ping
Lina Ping
XLXiangnan Liu
Xiangnan Liu
Extracting fragmented cropland is essential for effective cropland management and sustainable agricultural development. However, extracting fragmented cropland presents significant challenges due to its irregular and blurred boundaries, as well as the diversity in crop types and distribution. Deep learning methods are widely used for land cover classification. This paper proposes ConvNeXt-U, a lightweight deep learning network that efficiently extracts fragmented cropland while reducing computational requirements and saving costs. ConvNeXt-U retains the U-shaped structure of U-Net but replaces the encoder with a simplified ConvNeXt architecture. The decoder remains unchanged from U-Net, and the lightweight CBAM (Convolutional Block Attention Module) is integrated. This module adaptively adjusts the channel and spatial dimensions of feature maps, emphasizing key features and suppressing redundant information, which enhances the capture of edge features and improves extraction accuracy. The case study area is Hengyang County, Hunan Province, China, using GF-2 remote sensing imagery. The results show that ConvNeXt-U outperforms existing methods, such as Swin Transformer (Acc = 85.1%, IoU = 79.1%), MobileNetV3 (Acc = 83.4%, IoU = 77.6%), VGG16 (Acc = 80.5%, IoU = 74.6%), and ResUnet (Acc = 81.8%, IoU = 76.1%), achieving an IoU of 79.5% and Acc of 85.2%. Under the same conditions, ConvNeXt-U has a faster inference speed of 37 images/s, compared to 28 images/s for Swin Transformer, 35 images/s for MobileNetV3, and 0.43 and 0.44 images/s for VGG16 and ResUnet, respectively. Moreover, ConvNeXt-U outperforms other methods in processing the boundaries of fragmented cropland, producing clearer and more complete boundaries. The results indicate that the ConvNeXt and CBAM modules significantly enhance the accuracy of fragmented cropland extraction. ConvNeXt-U is also an effective method for extracting fragmented cropland from remote sensing imagery.

12 December 2025 in Remote Sensing
Verified
Journal Article
Open access
PDF file
YYYujuan Yang
Yujuan Yang
SCShi Cao
Shi Cao
XLXia Lu
Xia Lu
LPLina Ping
Lina Ping
XFXiang Fan
Xiang Fan
XLXiangnan Liu
Xiangnan Liu
Monitoring and identifying perennial cash crops is essential for optimizing agricultural resource allocation and supporting sustainable rural development. However, cross-regional recognition remains challenging due to cloud contamination, irregular mountainous topography, and climatic-driven phenological shifts. To address these issues, we propose a Region-Adaptive Multi-Head Phenology-Aware Network (RAM-PAMNet) that incorporates three key innovations. First, a Multi-source Temporal Attention Fusion (MTAF) module dynamically fuses Sentinel-1 SAR and Sentinel-2 optical time series to enhance temporal consistency and cloud robustness. Second, a Region-Aware Module (RAM) encodes topographic and climatic factors to adaptively adjust phenological windows across regions. Third, a Multi-Head Phenology-Aware Module (MHA-PAM) captures short-, mid-, and long-term phenological rhythms while integrating region-modulated attention for adaptive feature learning. The model was trained and validated in Changde, Hunan (694 patches; augmented to 2776; 70%/15%/15% split) and independently tested in Yaan, Sichuan (574 patches), two regions with contrasting elevation, terrain complexity, and hydrothermal regimes. RAM-PAMNet achieved an OA of 83.3%, mean F1 of 78.8%, and mIoU of 65.4% in Changde, and maintained strong generalization in Yaan with an mIoU of 59.2% and a DecayRate of 9.5, outperforming all baseline models. These results demonstrate that RAM-PAMNet effectively mitigates regional phenological mismatches and improves perennial crop mapping across heterogeneous environments. The proposed framework provides an interpretable and region-adaptive solution for large-scale monitoring of tea, citrus, and grape.

23 January 2026 in Remote Sensing
Verified
Journal Article
Open access
PDF file
TSTong Shi
Tong Shi
SCShi Cao
Shi Cao
XLXia Lu
Xia Lu
LPLina Ping
Lina Ping
XFXiang Fan
Xiang Fan
XLXiangnan Liu
Xiangnan Liu
Camellia oleifera, a woody oil crop unique to China, plays a crucial role in alleviating the global pressure of edible oil supply and maintaining ecological security. However, it remains challenging to accurately forecast Camellia oleifera yield in complex landscapes using only remote sensing data. The aim of this study is to develop an interpretable deep learning model, namely Shapley Additive Explanations–guided Attention–long short-term memory (SALSTM), for estimating Camellia oleifera yield by integrating an improved spectral bloom index and environmental parameters. The study area is located in Hengyang City in Hunan Province. Sentinel-2 imagery, meteorological observation from 2019 to 2023, and topographic data were collected. First, an improved spectral bloom index (ISBI) was constructed as a proxy for flowering density, while average temperature, precipitation, accumulated temperature, and wind speed were selected to represent environmental regulation variables. Second, a SALSTM model was designed to capture temporal dynamics from multi-source inputs, in which the LSTM module extracts time-dependent information and an attention mechanism assigns time-step-wise weights. Feature-level importance derived from SHAP analysis was incorporated as a guiding prior to inform attention distribution across variable dimensions, thereby enhancing model transparency. Third, model performance was evaluated using root mean square error (RMSE) and coefficient of determination (R2). The result show that the constructed SALSTM model achieved strong predictive performance in predicting Camellia oleifera yield in Hengyang City (RMSE = 0.5738 t/ha, R2 = 0.7943). Feature importance analysis results reveal that ISBI weight > 0.26, followed by average temperature and precipitation from flowering to fruit stages, these features are closely associated with C. oleifera yield. Spatially, high-yield zones were mainly concentrated in the central–southern hilly regions throughout 2019–2023, In contrast, low-yield zones were predominantly distributed in the northern and western mountainous areas. Temporally, yield hotspots exhibited a gradual increasing while low-yield zones showed mild fluctuations. This framework provides an effective and transferable approach for remote sensing-based yield estimation of flowering and fruit-bearing crops in complex landscapes.

6 December 2024 in Land
Verified
Journal Article
Open access
PDF file
YMYunong Ma
Yunong Ma
SCShi Cao
Shi Cao
XLXia Lu
Xia Lu
JPJiqing Peng
Jiqing Peng
LPLina Ping
Lina Ping
MLMeiling Liu
Meiling Liu
Analyzing farmland landscape pattern variations induced by human activities can support effective decision making by governments to improve land use efficiency. However, research on long-term and continuous spatial process changes in farmland is scarce, and spatial pattern changes in farmlands remain insufficiently understood. Moreover, studies in which researchers have utilized dynamic process analysis to describe farmlands are relatively limited. This study aimed to apply the state-and-evolution detection model (SEDM), generated from long-term remote sensing data, to characterize farmland spatial pattern variations in Hengyang City, Hunan Province. Annual farmland data from 1990 to 2022, change type samples, and auxiliary data were collected, and six types of spatial pattern variations (perforation, dissection, shrinkage, creation, enlargement, and aggregation) were defined for the study area. Subsequently, the SEDM was applied based on four landscape indices. Finally, spatiotemporal evolution features, namely evolution times, evolution duration, and dominant evolution pattern, were quantified. The farmland in the study area exhibited a generally upward trend with fluctuations. The maximum area was followed by shrinkage (S), perforation (P), and enlargement (E). In over 70% of the study area, fewer than three evolution times occurred over three decades. The dominant evolution patterns were P–S, S–P, and E–P for single evolution events, and P–S–P, S–P–S, and P–S–S for double events. The model achieved an overall accuracy of 85%, thus demonstrating its effectiveness in characterizing landscape pattern variations and providing valuable insights for researchers and policy makers to develop strategies for farmland protection.

16 April 2025 in Polish Journal of Environmental Studies
Verified
Journal Article
Open access
SCShi Cao
Shi Cao
XLXia Lu
Xia Lu
LPLina Ping
Lina Ping
XFXiang Fan
Xiang Fan
YZYuxin Zhao
Yuxin Zhao
Ling WuLing Wu
Ling Wu
Most change detection algorithms are designed to detect specific forest disturbances, which may not effectively capture diverse events. These algorithms typically model seasonal changes using dense observations to reduce phenological noise, making them unsuitable for regions with frequent cloud cover. In this study, we propose a change detection algorithm for diverse forest disturbances of varying magnitudes using sparse Landsat time series. The Normalized Difference Moisture Index (NDMI) was spatially normalized (SNDMI) to remove forest seasonality, reducing the need for dense observations. Residuals obtained from SNDMI fitting using the spatial error model were input into the Exponentially Weighted Moving Average t (EWMA-t) chart, designed for sparse data and sensitive to low-magnitude disturbances. An adaptive strategy to the EWMA-t chart (AEWMA-t) that adjusts the weights of historical chart values and current residual statistics is introduced. Low-magnitude disturbances exceed control limits with small values, while high-magnitude disturbances prioritize current residuals for rapid detection. Disturbances are identified when chart values consecutively exceed control limits. Applied to a cloudy subtropical forest with diverse disturbances, the proposed algorithm achieved 84.6% and 89.3% accuracy in spatial and temporal domains, offering a reliable approach for detecting diverse disturbances in low-data regions.