Content of Technology and application of smart forestry and grassland in our journal

Published in last 1 year |  In last 2 years |  In last 3 years |  All Please wait a minute... For Selected: Download Citations EndNote Reference Manager ProCite BibTeX RefWorks Toggle Thumbnails Select Wildlife Image Recognition of Infrared Cameras in Beijing Area Based on an Improvement ConvNeXt Model Jiandong Qi,Shangzi Zheng,Ziyi Chen,Zhongtian Ma Scientia Silvae Sinicae    2024, 60 (8): 33-45.   DOI: 10.11707/j.1001-7488.LYKX20230276 Abstract （97）   HTML （17）    PDF （3360KB）（94）       Knowledge map    Save Objective: Aiming at the problems of large amount of data, high proportion of invalid images, and complex image backgrounds in wild animal images captured by infrared cameras, a model that can automatically and accurately recognize images is proposed, providing more efficient support for biodiversity research and wildlife conservation work. Method: Collect and organize approximately 5 TB of image data captured by infrared cameras at various stations of the Beijing Ecological Observatory Network over the past 4 years. After manual annotation and data augmentation, create a total of 4234 image datasets in 10 categories. Based on ConvNeXt convolutional neural network and combined with the characteristics of wild animal image datasets in Beijing, a BSGG-ConvNeXt model was designed. BlurPool, SENet, global response normalization layer (GRN), and GCNet were used to improve the recognition ability of the model. The impact of training strategies on the recognition accuracy of ConvNeXt network was explored on a self-built dataset. By comparing with other classic models, the advantages of the BSGG-ConvNeXt model are clarified. Verify the generalization ability of the model using publicly available infrared wildlife snapshot serengeti (SS) dataset andcaltech camera traps (CCT) dataset. Result: Taking the ConvNeXt size model of the ConvNeXt model as an example, the accuracy in the self-built dataset is 74.13%, and the multiply add cumulative operands (MACs) are 4.47×109. By applying different improvement schemes, it was found that the accuracy increased by 2.2% and MACs decreased to 1.07×109 after using BlurPool. After using SENet, the accuracy improved by 3.2%. After using GRN and removing the scaling layer, the accuracy improved to 87.18% and the number of parameters increased to 27.88×106. After using GCNet, the accuracy was improved to 75.44% without increasing the computational load, but the number of parameters increased to 28.25×106. The BSGG-ConvNeXt obtained by combining the above improvement schemes is applied to the ConvNeXt-T model to obtain the BSGG-ConvNeXt-T model. Although there is a slight increase in the number of parameters, the MACs are reduced to 1.07×109, and the accuracy of the model is improved to 83.63%, which is higher than the original model. After using pre-trained weights, the accuracy of the BSGG-ConvNeXt-T model can reach 94.07%, which is higher than the accuracy of ResNet-50 (76.39%), ResNeXt-50 (87.60%), MobileViT (90.00%), DenseNet (87.66%), RegNet (69.90%), ConvNeXtv2 (91.93%), SwinTransformer (86.23%), and MobileOne (71.53%) models. After applying the BSGG-ConvNeXt model to four different network sizes of ConvNeXt models, its performance in the self-built dataset is better than that of the unimproved model. The recognition accuracy of the BSGG-ConvNeXt model in the SS dataset can reach 50.28%, and the recognition accuracy in the CCT dataset can reach 56.15%, both of which are higher than the accuracy of the original model. Conclusion: The BSGG-ConvNeXt model has a higher accuracy in recognizing wild animal images captured by infrared cameras, and performs well on both self built and publicly available wild animal infrared image datasets, with a certain degree of generalization ability. Table and Figures | Reference | Related Articles | Metrics Select Predicting the Distribution of Suitable Habitats for Oriental Storks Based on Satellite Tracking in Yellow River Delta Lixia Chen,Feng Lu,Hongxing Jiang,Ge Sun,Xiupeng Yue,Yixuan Wang,Tong Gao,Xingbo Hu,Changqing Ding Scientia Silvae Sinicae    2024, 60 (8): 46-56.   DOI: 10.11707/j.1001-7488.LYKX20230642 Abstract （99）   HTML （16）    PDF （1746KB）（280）       Knowledge map    Save Objective: Based on locations tracked by satelite of Oriental Stork born in the Yellow River Delta, the nation-wide distributions of suitable habitats in both summer and winter were predicted. This study was intended to provide an important scientific basis for the collaborative conservation of Oriental Stork. Method: Using the satellite tracking data of 80 Oriental Storks in the Yellow River Delta of Shandong Province between 2016 and 2022, 5 283 habitat locations in summer and 4 730 habitat locations in winter were selected through spatial auto-correlation analysis. The MaxEnt model was employed to predict the potential suitable habitat distribution for Oriental Storks in both summer and winter, and to rank the suitability habitats. Result: The analysis revealed that the MaxEnt predictions are satisfactory, with AUC values exceeding 0.8 for both the summer (0.866) and winter (0.838). Six environmental factors such as elevation, isotherm, distance to lakes and rivers, average daily temperature range, precipitation seasonality and the land-use type—significantly influence the distribution of Oriental Storks in the Yellow River Delta. In summer, there was a highly suitable habitat area of 14.75×104 km2 for Oriental Storks, constituting 13.01% of the total suitable area. These habitats were predominantly distributed in the Songnen Plain (57.52%), the Bohai Bay (17.50%), the Liaohe Plain (14.94%). The main habitat types in summer included wetlands (35.30%), cropland (33.74%), water (12.43%). The highly suitable winter habitat for Oriental Storks covered an area of 11.63×104 km2, constituting 11.62% of the total suitable area. It was primarily situated in the Middle-Lower Yangtze Plains (49.02%), Hongze Lake of the Huaihe River basin (23.75%), the Bohai Bay (11.52%). The main winter habitats included wetlands (40.56%), cropland (29.68%), water (18.45%). The overlapping area of highly suitable habitat in summer and winter was 2.13×104 km2, primarily concentrated in the Bohai Bay (59.81%), the coastal area of Shandong Peninsula (23.24%) and Jiangsu (9.34%). Conclusion: The results indicate that the suitable habitat in summer extends to Songnen Plain, Sanjiang Plain and Liaohe Plain, challenging conventional understanding about the fledging dispersal for Oriental Storks. for Oriental Storks. This study offers a crucial scientific foundation for the collaborative conservation of this species. Table and Figures | Reference | Related Articles | Metrics Select Wildlife Images Recognition Method Based on Wasserstein Distance and Correlation Alignment Transfer Learning Changchun Zhang,Dafang Li,Junguo Zhang Scientia Silvae Sinicae    2024, 60 (8): 25-32.   DOI: 10.11707/j.1001-7488.LYKX20230399 Abstract （99）   HTML （14）    PDF （2546KB）（119）       Knowledge map    Save Objective: This study aims to address the influence of complex factors such as lighting, background, and shooting scale on the accuracy of wildlife image recognition. Method: In this study, the wild animal images captured by infrared triggered cameras in the wild were used as the object: 1) Two publicly available wildlife datasets, ENA24 and NACTI, were used to construct disjoint datasets S1 and S2, comprising a total of 11 animal categories and 25 591 images. 2) To tackle domain shift issues, a ResNet50 network was utilized as a feature extraction module to build a domain adversarial network, effectively alleviating domain bias. 3) A representation learning network incorporating Wasserstein distance and correlation alignment was proposed to establish a transfer learning network for feature extraction and recognition, so as to further exploit transferable features. Result: The performance of different models in wildlife recognition was evaluated using the average accuracy metric. Results indicated that the average accuracy on 11 wildlife categories for eight models, namely ResNet50, DDC, DCORAL, DAN, DANN, CDAN, HAN, and JTN, was 48.4%, 51.6%, 49.6%, 52.6%, 45.2%, 50.9%, 54.6%, and 53.5%, respectively. Upon enhancing the ResNet50 base model with improved residual modules and introducing a representation learning network incorporating Wasserstein distance and correlation alignment, the average accuracy for 11 wildlife categories was improved by 2.7% compared to the existing best result with the comparative methods. Conclusion: The transfer learning method based on Wasserstein distance and correlation alignment has achieved an average accuracy of 57.3% in wildlife recognition. The introduction of representation learning based on Wasserstein distance and correlation alignment can effectively improve the accuracy of the wildlife recognition model. Table and Figures | Reference | Related Articles | Metrics Select Individual Tree Segmentation of UAV-LiDAR Based on the Combination of CHM and DSM Zhongyang Hu,Liang Shan,Xiangyu Chen,Kunyong Yu,Jian Liu Scientia Silvae Sinicae    2024, 60 (8): 14-24.   DOI: 10.11707/j.1001-7488.LYKX20230079 Abstract （186）   HTML （45）    PDF （6244KB）（171）       Knowledge map    Save Objective: Considering that the unmanned aerial vehicle(UAV) LiDAR-derived canopy height model (CHM) is prone to distortion in areas with complex terrain, which significantly limits the accuracy of the individual tree segmentation, this study aims to propose a new method that utilizes the combination of CHM and digital surface model (DSM) to segment individual tree. Method: By using UAV-LiDAR data and choosing 3 standard plots of middle-aged forest and young forest in steep areas of Cunninghamia lanceolata plantation with middle and high crown density in Yangkou forest farm, Shunchang country, Fujian Province, this paper used the field-measured and visual interpretation data to calculate and analyze the accuracy of treetop detection and individual tree segmentation which were acquired by using local maximum algorithm with 4 fixed window size and extremum controlling watershed algorithm based on the combination of CHM and DSM and compared them with the traditional method of using CHM alone. Result: In the matter of treetop detection, number of detecting treetops and detection percentage for each plot showed a downward trend with window size increasing; The optimal window size of 3 middle-aged forest plots is 0.3 m and the optimal window size of 3 young forest plots is 0.2 m. In this configuration, the individual tree of one-to-one correspondence and the producer’s accuracy are the maxima. By using the CHM alone, the detection percentage are higher but the accuracy of treetop detection are lower than using the combination of CHM and DSM because the CHM-based local maximum method is prone to multiple detecting treetops in the individual tree on the optimal window size. The accuracy of treetop detection in the young forest plots are higher than in the middle-aged forest plots. The reason is that the crown breadth and adjacent distance of individual trees in the young forest plots are more consistent which is more adapted to the local maximum algorithm. In the matter of individual tree segmentation, on the optimal window size, by using the combination of CHM and DSM, the accuracy of individual tree segmentation are higher than using the CHM alone and the accuracy of individual tree segmentation in the young forest plots are higher than the middle-aged forest plots. Conclusion: Because the direct data source of treetop detection and individual tree segmentation is DSM which reflects the crown surface veritably with no distortion, the result represents the real individual tree crown boundary and reaches the high accuracy in the young and middle ages of Cunninghamia lanceolata plantation with middle and high crown density (the detection rate r is more than 88%, the accuracy p is more than 92% and the F-score is more than 91% for each plot). Moreover, in this study, the method integrates into the ArcGIS Model Builder which is helpful to build an accurate, automatic, and integrated platform for individual tree segmentation of UAV-LiDAR. Table and Figures | Reference | Related Articles | Metrics Select Spatiotemporal Patterns and Driving Forces of Vegetation Restoration and Degradation in Dongting Lake Wetland Yutian Zhang,Junnan Shi,Huaiqing Zhang,Binglun Wu Scientia Silvae Sinicae    2024, 60 (8): 1-13.   DOI: 10.11707/j.1001-7488.LYKX20240048 Abstract （230）   HTML （27）    PDF （2621KB）（211）       Knowledge map    Save Objective: Wetland vegetation can purify the environment, regulate climate, and improve the soil, which is of great significance to the ecological security and stability of the wetland system. Wetland vegetation communities are, however, under serious threat from accelerated global warming and human disturbance. Probing the long-term spatial-temporal pattern of wetland vegetation cover change and its response to climate change and human activities are important for informing decisions on wetland protection. Method: It is difficult to collect long-term, reliable optical observations with high spatiotemporal resolution in the Dongting Lake wetland due to its location in the subtropical monsoon climate zone and its frequent cloud and rainy conditions. Firstly, this study used the flexible spatiotemporal data analysis fusion (FSDAF) algorithm to fuse Landsat and MODIS images to obtain the monthly normalized difference vegetation index (NDVI) time series during the study period (2000—2019). To quantify the response of wetland vegetation cover to environmental changes and human disturbance at different time scales, the seasonal and trend components of wetland vegetation NDVI in Dongting Lake were separated by an improved seasonal-trend decomposition procedure based on the loess (STL) time series decomposition method. Based on the linear trend analysis and NDVI seasonal and trend component data with a high spatiotemporal resolution, the spatiotemporal dynamic patterns of wetland vegetation at different scales were identified. Finally, a partial correlation-based approach was used to quantitatively assess the contributions of three major climatic factors (i.e., temperature, precipitation, and solar radiation) and anthropogenic factors to seasonal and trend vegetation changes from 2000 to 2019. Result: 1) From 2000 to 2019, the trend component and seasonal component of NDVI vegetation in Dongting Lake wetland showed spatial differentiation patterns, but overall exhibited a “greening” trend, with a change rate of 4.8×10?3 a?1 and 0.4×10?3 a?1, respectively. 2) Vegetation growth has a significant positive correlation with temperature and solar radiation, and its seasonal correlation with vegetation change is generally greater than trend correlation. The correlation between precipitation and vegetation is relatively low, and it has a negative correlation with rice (trend component R=?0.27; seasonal component R=?0.42). 3) From 2000 to 2019, human factors and climate change have driven 58% and 42% of the NDVI vegetation changes in the Dongting Lake wetland, respectively. Among them, the relative contribution rates of human factors to the long-term and seasonal growth and restoration of wetland vegetation are 55% and 62%, respectively, while the relative contribution rates of climate change to the long-term and seasonal degradation of wetland vegetation are 53% and 56%, respectively. Conclusion: Human factors have promoted the growth of wetland vegetation and are the main driving force behind the greening of Dongting Lake wetland vegetation from 2000 to 2019, while climate change poses a threat to wetland ecosystems. Adopting appropriate ecological protection and restoration measures is still an important means to achieve sustainable development of the Dongting Lake wetland ecosystem in the future. This study elucidates the response mechanism of Dongting Lake wetland vegetation to climate change and human activities, providing a scientific basis for spatial decision-making on wetland conservation. Table and Figures | Reference | Related Articles | Metrics