YOLOv4-Lite–Based Urban Plantation Tree Detection and Positioning With High-Resolution Remote Sensing Imagery

Automatic tree identification and position using high-resolution remote sensing images are critical for ecological garden planning, management, and large-scale environmental quality detection. However, existing single-tree detection methods have a high rate of misdetection in forests not only due to the similarity of background and crown colors but also because light and shadow caused abnormal crown shapes, resulting in a high rate of misdetections and missed detection. This article uses urban plantations as the primary research sample. In conjunction with the most recent deep learning method for object detection, a single-tree detection method based on the lite fourth edition of you only look once (YOLOv4-Lite) was proposed. YOLOv4’s object detection framework has been simplified, and the MobileNetv3 convolutional neural network is used as the primary feature extractor to reduce the number of parameters. Data enhancement is performed for categories with fewer single-tree samples, and the loss function is optimized using focal loss. The YOLOv4-Lite method is used to detect single trees on campus, in an orchard, and an economic plantation. Not only is the YOLOv4-Lite method compared to traditional methods such as the local maximum value method and the watershed method, where it outperforms them by nearly 46.1%, but also to novel methods such as the Chan-Vese model and the template matching method, where it outperforms them by nearly 26.4%. The experimental results for single-tree detection demonstrate that the YOLOv4-Lite method improves accuracy and robustness by nearly 36.2%. Our work establishes a reference for the application of YOLOv4-Lite in additional agricultural and plantation products.

Urban Forest Ecosystem Services Vary with Land Use and Species: A Case Study of Kyoto City

The demand for urban ecosystem services increases with the rapid growth of the urban population. The urban forest is a crucial ecosystem services provider in cities. To achieve a better estimation of urban ecosystem services, an understanding of the link between heterogeneity and ecosystem services within cities is needed. Other than street trees and forest remnants, the contribution of dispersed green spaces should also be considered. In this study, a ground-based sample quadrat investigation of trees across a sequence of land types in Kyoto City was applied. The ecosystem services and monetary values of trees were further calculated using a customized i-Tree Eco tool. The ecosystem services calculated include carbon storage and sequestration, air pollutants removal, and runoff reduction. Ecosystem services of different land use classes were compared at both quadrat and single-tree levels. We found no significant difference across land use for all the ecosystem services at the quadrat level. However, ecosystem services were different across land use at the single-tree level. We performed a species-specific analysis and found that the pattern of ecosystem services at the single-tree level across land use varies with both the service tested and species. Our study suggests that the heterogeneity within a city should be considered when estimating urban ecosystem services. The results also provide insight into the urban green space management of Kyoto City.

Single Shot MultiBox Detector for Urban Plantation Single Tree Detection and Location With High-Resolution Remote Sensing Imagery

Using high-resolution remote sensing images to automatically identify individual trees is of great significance to forestry ecological environment monitoring. Urban plantation has realistic demands for single tree management such as catkin pollution, maintenance of famous trees, landscape construction, and park management. At present, there are problems of missed detection and error detection in dense plantations and complex background plantations. This paper proposes a single tree detection method based on single shot multibox detector (SSD). Optimal SSD is obtained by adjusting feature layers, optimizing the aspect ratio of a preset box, reducing parameters and so on. The optimal SSD is applied to single tree detection and location in campuses, orchards, and economic plantations. The average accuracy based on SSD is 96.0, 92.9, and 97.6% in campus green trees, lychee plantations, and palm plantations, respectively. It is 11.3 and 37.5% higher than the latest template matching method and chan-vese (CV) model method, and is 43.1 and 54.2% higher than the traditional watershed method and local maximum method. Experimental results show that SSD has a strong potential and application advantage. This research has reference significance for the application of an object detection framework based on deep learning in agriculture and forestry.

Estimation of Individual Tree Stem Biomass in an Uneven-Aged Structured Coniferous Forest Using Multispectral LiDAR Data

Stem biomass is a fundamental component of the global carbon cycle that is essential for forest productivity estimation. Over the last few decades, Light Detection and Ranging (LiDAR) has proven to be a useful tool for accurate carbon stock and biomass estimation in various biomes. The aim of this study was to investigate the potential of multispectral LiDAR data for the reliable estimation of single-tree total and barkless stem biomass (TSB and BSB) in an uneven-aged structured forest with complex topography. Destructive and non-destructive field measurements were collected for a total of 67 dominant and co-dominant Abies borisii-regis trees located in a mountainous area in Greece. Subsequently, two allometric equations were constructed to enrich the reference data with non-destructively sampled trees. Five different regression algorithms were tested for single-tree BSB and TSB estimation using height (height percentiles and bicentiles, max and average height) and intensity (skewness, standard deviation and average intensity) LiDAR-derived metrics: Generalized Linear Models (GLMs), Gaussian Process (GP), Random Forest (RF), Support Vector Regression (SVR) and Extreme Gradient Boosting (XGBoost). The results showcased that the RF algorithm provided the best overall predictive performance in both BSB (i.e., RMSE = 175.76 kg and R2 = 0.78) and TSB (i.e., RMSE = 211.16 kg and R2 = 0.65) cases. Our work demonstrates that BSB can be estimated with moderate to high accuracy using all the tested algorithms, contrary to the TSB, where only three algorithms (RF, SVR and GP) can adequately provide accurate TSB predictions due to bark irregularities along the stems. Overall, the multispectral LiDAR data provide accurate stem biomass estimates, the general applicability of which should be further tested in different biomes and ecosystems.

Evolutionary Computation for Feature Manipulation in Classification on High-dimensional Data

<p>More and more high-dimensional data appears in machine learning, especially in classification tasks. With thousands of features, these datasets bring challenges to learning algorithms not only because of the curse of dimensionality but also the existence of many irrelevant and redundant features. Therefore, feature selection and feature construction (or feature manipulation in short) are essential techniques in preprocessing these datasets. While feature selection aims to select relevant features, feature construction constructs high-level features from the original ones to better represent the target concept. Both methods can decrease the dimensionality and improve the performance of learning algorithms in terms of classification accuracy and computation time. Although feature manipulation has been studied for decades, the task on high-dimensional data is still challenging due to the huge search space. Existing methods usually face the problem of stagnation in local optima and/or require high computation time. Evolutionary computation techniques are well-known for their global search. Particle swarm optimisation (PSO) and genetic programming (GP) have shown promise in feature selection and feature construction, respectively. However, the use of these techniques to high-dimensional data usually requires high memory and computation time. The overall goal of this thesis is to investigate new approaches to using PSO for feature selection and GP for feature construction on high-dimensional classification problems. This thesis focuses on incorporating a variety of strategies into the evolutionary process and developing new PSO and GP representations to improve the effectiveness and efficiency of PSO and GP for feature manipulation on high-dimensional data. This thesis proposes a new PSO based feature selection approach to high-dimensional data by incorporating a new local search to balance global and local search of PSO. A hybrid of wrapper and filter evaluation method which can be sped up in the local search is proposed to help PSO achieve better performance, scalability and robustness on high-dimensional data. The results show that the proposed method significantly outperforms the compared methods in 80% of the cases with an increase up to 16% average accuracy while reduces the number of features from one to two orders of magnitude. This thesis develops the first PSO based feature selection via discretisation method that performs both multivariate discretisation and feature selection in a single stage to achieve better solutions than applying these techniques separately in two stages. Two new PSO representations are proposed to evolve cut-points for multiple features simultaneously. The results show that the proposed method selects less than 4.6% of the features in all cases to improve the classification performance from 5% to 23% in most cases. This thesis proposes the first clustering-based feature construction method to improve the performance of single-tree GP on high-dimensional data. A new feature clustering method is proposed to automatically group similar features into the same group based on a given redundancy level. The results show that compared with standard GP, the new method can select less than half of the features to construct a new high-level feature that achieves significantly better accuracy in most cases. The combination of the single constructed feature and the selected ones achieves the best performance among different feature sets created from a single tree. This thesis develops the first class-dependent multiple feature construction method using multi-tree GP for high-dimensional data. A new GP representation and a new filter fitness function that combines two filter measures are proposed to evaluate the whole set of constructed features more effectively and efficiently. The results show that in 83% of the cases, with less than 10 constructed features, the class-dependent method increases up to 32% average accuracy on using all the original thousands of features and 10% on using those constructed by the class-independent method.</p>

Evolutionary Computation for Feature Manipulation in Classification on High-dimensional Data

<p>More and more high-dimensional data appears in machine learning, especially in classification tasks. With thousands of features, these datasets bring challenges to learning algorithms not only because of the curse of dimensionality but also the existence of many irrelevant and redundant features. Therefore, feature selection and feature construction (or feature manipulation in short) are essential techniques in preprocessing these datasets. While feature selection aims to select relevant features, feature construction constructs high-level features from the original ones to better represent the target concept. Both methods can decrease the dimensionality and improve the performance of learning algorithms in terms of classification accuracy and computation time. Although feature manipulation has been studied for decades, the task on high-dimensional data is still challenging due to the huge search space. Existing methods usually face the problem of stagnation in local optima and/or require high computation time. Evolutionary computation techniques are well-known for their global search. Particle swarm optimisation (PSO) and genetic programming (GP) have shown promise in feature selection and feature construction, respectively. However, the use of these techniques to high-dimensional data usually requires high memory and computation time. The overall goal of this thesis is to investigate new approaches to using PSO for feature selection and GP for feature construction on high-dimensional classification problems. This thesis focuses on incorporating a variety of strategies into the evolutionary process and developing new PSO and GP representations to improve the effectiveness and efficiency of PSO and GP for feature manipulation on high-dimensional data. This thesis proposes a new PSO based feature selection approach to high-dimensional data by incorporating a new local search to balance global and local search of PSO. A hybrid of wrapper and filter evaluation method which can be sped up in the local search is proposed to help PSO achieve better performance, scalability and robustness on high-dimensional data. The results show that the proposed method significantly outperforms the compared methods in 80% of the cases with an increase up to 16% average accuracy while reduces the number of features from one to two orders of magnitude. This thesis develops the first PSO based feature selection via discretisation method that performs both multivariate discretisation and feature selection in a single stage to achieve better solutions than applying these techniques separately in two stages. Two new PSO representations are proposed to evolve cut-points for multiple features simultaneously. The results show that the proposed method selects less than 4.6% of the features in all cases to improve the classification performance from 5% to 23% in most cases. This thesis proposes the first clustering-based feature construction method to improve the performance of single-tree GP on high-dimensional data. A new feature clustering method is proposed to automatically group similar features into the same group based on a given redundancy level. The results show that compared with standard GP, the new method can select less than half of the features to construct a new high-level feature that achieves significantly better accuracy in most cases. The combination of the single constructed feature and the selected ones achieves the best performance among different feature sets created from a single tree. This thesis develops the first class-dependent multiple feature construction method using multi-tree GP for high-dimensional data. A new GP representation and a new filter fitness function that combines two filter measures are proposed to evaluate the whole set of constructed features more effectively and efficiently. The results show that in 83% of the cases, with less than 10 constructed features, the class-dependent method increases up to 32% average accuracy on using all the original thousands of features and 10% on using those constructed by the class-independent method.</p>

3D Point Cloud of Single Tree Branches and Leaves Semantic Segmentation Based on Modified PointNet Network

Semantic segmentation of single tree 3D point cloud is one of the key technologies in building tree model. It plays an important role in tree skeleton extraction, tree pruning, tree model reconstruction and other fields. Because the area of a single leaf is smaller than that of the whole tree, the segmentation of branches and leaves is a challenging problem. In view of the above problems, this paper first migrates PointNet to the tree branch and leaf point cloud segmentation, and proposes an automatic segmentation method based on improved PointNet. According to the difference of normal direction between leaves and branches, the point cloud information of three dimensions coordinates, color and normal vector is input into the point feature space. In data processing, increase the number of each block data, so that the network can better learn features. MLP is added to the original PointNet network to improve the ability of extracting and learning local features. In addition, in the process of feature extraction, jump connection is added to realize feature reuse and make full use of different levels of features. The original 1×1 filter of PointNet is replaced by 3×1 filter to improve the segmentation accuracy of tree point cloud. The focus loss function focal loss is introduced into the field of 3D point cloud to reduce the impact of the imbalance of point cloud samples on the results. The results show that the improved method improves the accuracy of tree branch point cloud segmentation compared with the original PointNet for branch and leaf segmentation. The segmentation accuracy of structural elements of branches and leaves is more than 88%, and MIoU is 48%.