分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2025-07-17 合作期刊: 《干旱区科学》
HUANG Cheng
HOU Shengtong
WANG Bao
SONG Yuchuan
Aikeremu ABULATIJIANG
MIN Jiuzhou
SHENG Jiandong
JIANG Ping'an
WANG Ze
CHENG Junhui
摘要： Biochar and animal manure application can improve crop yields in salt-affected soil. Previous studies have primarily applied biochar and animal manure either alone or at fixed ratios, while their combined effects with varying combination proportions are still unclear. To address this knowledge gap, we performed a 2-a experiment (2023–2024) in a salinized cotton field in Wensu County of Xinjiang Uygur Autonomous Region of China with the following 6 treatments: control; application of biochar (10 t/hm2) alone (BC100%); application of cow manure (10 t/hm2) alone (CM100%); application of 70% biochar (7 t/hm2) combined with 30% cow manure (3 t/hm2) (BC70%+CM30%); application of 50% biochar (5 t/hm2) combined with 50% cow manure (5 t/hm2) (BC50%+CM50%); and application of 30% biochar (3 t/hm2) combined with 70% cow manure (7 t/hm2) (BC30%+CM70%). By measuring soil pH, electrical conductivity, soil organic matter, available phosphorus, available potassium, and available nitrogen at 0–20 and 20–40 cm depths, as well as yield components and cotton yield in 2023 and 2024, this study revealed that soil nutrients in the 0–20 cm depth were more sensitive to the treatment. Among all the treatments, BC50%+CM50% treatment had the highest value of soil pH (9.63±0.07) but the lowest values of electrical conductivity (161.9±31.8 μS/cm), soil organic matter (1.88±0.27 g/kg), and available potassium (42.72±8.25 mg/kg) in 2024. Moreover, the highest cotton yield (5336.63±467.72 kg/hm2) was also observed under BC50%+CM50% treatment in 2024, which was 1.9 times greater than that under the control treatment. In addition, cotton yield in 2023 was jointly determined by yield components (density and number of cotton bolls) and soil nutrients (available phosphorus and available potassium), but in 2024, cotton yield was only positively related to yield components (density, number of cotton bolls, and single boll weight). Overall, this study highlighted that in salt-affected soil, the combination of biochar and cow manure at a 1:1 ratio is recommended for increasing cotton yield and reducing soil salinity stress.
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2025-07-07 合作期刊: 《干旱区科学》
摘要： Cotton, as one of important economic crops, is widely planted in the saline-alkaline soil of southern Xinjiang, China. Moreover, in order to control the saline-alkaline content for seed germination and seedlings survive of cotton, farmers always adopt salt leaching during winter and spring seasons. However, excessive amount of salt leaching might result in the waste of water resources and unsuitable irrigation seasons might further increase soil salinization. In this study, a field experiment was conducted in the saline-alkaline soil in 2020 and 2021 to determine the effects of leaching amount and period on water-salinity dynamics and cotton yield. Five leaching amounts (0.0 (W0), 75.0 (W1), 150.0 (W2), 225.0 (W3), and 300.0 (W4) mm) and three leaching periods (seedling stage (P1), seedling and squaring stages (P2), and seedling, squaring, flowering, and boll setting stages (P3)) were used. In addition, a control treatment (CK) with a leaching amount of 300.0 mm in spring was performed. The soil water-salt dynamics, cotton growth, seed cotton yield, water productivity (WP), and irrigation water productivity (WPI) were analyzed. Results showed that leaching significantly decreased soil electrical conductivity (EC), and W3P2 treatment reduced EC by 11.79% in the 0–100 cm soil depth compared with CK. Plant height, stem diameter, leaf area index, and yield under W3 and W4 treatments were greater than those under W1 and W2 treatments. Compared with W3P1 and W3P3 treatments, seed cotton yield under W3P2 treatment significantly enhanced and reached 6621 kg/hm2 in 2020 and 5340 kg/hm2 in 2021. Meanwhile, WP and WPI under W3P2 treatment were significantly higher than those under other leaching treatments. In conclusion, the treatment of 225.0 mm leaching amount and seedling and squaring stages-based leaching period was beneficial for the salt control, efficient water utilization, and yield improvement of cotton in southern Xinjiang, China.
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2025-07-07 合作期刊: 《干旱区科学》
摘要： Integrating sprinkler with deficit irrigation system is a new approach to improve crop water productivity and ensure water and food security in arid areas of India. This study undertook a field experiment of sprinkler-irrigated cumin (variety GC-4) with a mini-lysimeter setup at an experimental research farm in Jodhpur, India during 2019–2022. Four irrigation treatments T1, T2, T3, and T4 were designed at irrigation water/cumulative pan evaporation (IW/CPE) of 1.0, 0.8, 0.6, and 0.4, respectively, with three replications. Daily actual crop evapotranspiration (ETc) was recorded and weekly soil moisture was monitored over the crop growth period. Quantities of applied water and drainage from mini-lysimeters were also measured at every irrigation event. Yield of cumin was recorded at crop maturity. Furthermore, change in farmer’s net income from 1-hm2 land was computed based on the cost of applying irrigation water and considering yield variations among the treatments. Results indicated the highest mean seasonal actual ETc (371.7 mm) and cumin yield (952.47 kg/hm2) under T1 (with full irrigation). Under T2, T3, and T4, the seasonal actual ETc decreased by 10.4%, 27.6%, and 41.3%, respectively, while yield declined by 5.0%, 28.4%, and 50.8%, respectively, as compared to the values under T1. Furthermore, crop water productivity of 0.272 (±0.068) kg/m3 under T2 was found relatively higher in comparison to other irrigation treatments, indicating that T2 can achieve improved water productivity of cumin in arid areas at an optimum level of deficit irrigation. The results of cost-economics indicated that positive change in farmer’s net income from 1-hm2 land was 108.82 USD under T2, while T3 and T4 showed net losses of 5.33 and 209.67 USD, respectively. Moreover, value of yield response factor and ratio of relative yield reductions to relative ETc deficits were found to be less than 1.00 under T2 (0.48), and more than 1.00 under T3 (1.07) and T4 (1.23). This finding further supports that T2 shows the optimized level of deficit irrigation that saves 20.0% of water with sacrificing 5.0% yield in the arid areas of India. Findings of this study provide useful strategies to save irrigation water, bring additional area under irrigation, and improve crop water productivity in India and other similar arid areas in the world.
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2025-02-21 合作期刊: 《干旱区科学》
摘要： Water is essential for agricultural production; however, climate change has exacerbated drought and water stress in arid and semi-arid areas such as Iran. Despite these challenges, irrigation water efficiency remains low, and current water management schemes are inadequate. Consequently, Iranian crops suffer from low water productivity, highlighting the urgent need for enhanced productivity and improved water management strategies. In this study, we investigated irrigation management conditions in the Hamidiyeh farm, Khuzestan Province, Iran and used the calibrated AquaCrop and WinSRFR (a surface irrigation simulation model) models to reflect these conditions. Subsequently, we examined different management scenarios using each model and evaluated the results from the second year. The findings demonstrated that combining simulation of the AquaCrop and WinSRFR models was highly effective and could be employed for irrigation management in the field. The AquaCrop model accurately simulated wheat yield in the first year, being 2.6 t/hm2, which closely aligned with the measured yield of 3.0 t/hm2. Additionally, using the WinSRFR model to adjust the length of existing borders from 200 to 180 m resulted in a 45.0% increase in efficiency during the second year. To enhance water use efficiency in the field, we recommended adopting borders with a length of 180 m, a width of 10 m, and a flow rate of 15 to 18 L/s. The AquaCrop and WinSRFR models accurately predicted border irrigation conditions, achieving the highest water use efficiency at a flow rate of 18 L/s. Combining these models increased farmers' average water consumption efficiency from 0.30 to 0.99 kg/m³ in the second year. Therefore, the results obtained from the AquaCrop and WinSRFR models are within a reasonable range and consistent with international recommendations. This adjustment is projected to improve the water use efficiency in the field by approximately 45.0% when utilizing the border irrigation method. Therefore, integrating these two models can provide comprehensive management solutions for regional farmers.
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2025-02-21 合作期刊: 《干旱区科学》
摘要： The Tarim River Basin (TRB) is a vast area with plenty of light and heat and is an important base for grain and cotton production in Northwest China. In the context of climate change, however, the increased frequency of extreme weather and climate events is having numerous negative impacts on the region's agricultural production. To better understand how unfavorable climatic conditions affect crop production, we explored the relationship of extreme weather and climate events with crop yields and phenology. In this research, ten indicators of extreme weather and climate events (consecutive dry days (CDD), min Tmax (TXn), max Tmin (TNx), tropical nights (TR), warm days (Tx90p), warm nights (Tn90p), summer days (SU), frost days (FD), very wet days (R95p), and windy days (WD)) were selected to analyze the impact of spatial and temporal variations on the yields of major crops (wheat, maize, and cotton) in the TRB from 1990 to 2020. The three key findings of this research were as follows: extreme temperatures in southwestern TRB showed an increasing trend, with higher extreme temperatures at night, while the occurrence of extreme weather and climate events in northeastern TRB was relatively low. The number of FD was on the rise, while WD also increased in recent years. Crop yields were higher in the northeast compared with the southwest, and wheat, maize, and cotton yields generally showed an increasing trend despite an earlier decline. The correlation of extreme weather and climate events on crop yields can be categorized as extreme nighttime temperature indices (TNx, Tn90p, TR, and FD), extreme daytime temperature indices (TXn, Tx90p, and SU), extreme precipitation indices (CDD and R95p), and extreme wind (WD). By using Random Forest (RF) approach to determine the effects of different extreme weather and climate events on the yields of different crops, we found that the importance of extreme precipitation indices (CDD and R95p) to crop yield decreased significantly over time. As well, we found that the importance of the extreme nighttime temperature (TR and TNx) for the yields of the three crops increased during 2005–2020 compared with 1990–2005. The impact of extreme temperature events on wheat, maize, and cotton yields in the TRB is becoming increasingly significant, and this finding can inform policy decisions and agronomic innovations to better cope with current and future climate warming.
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2024-02-21 合作期刊: 《干旱区科学》
摘要： Drip irrigation and flood irrigation are major irrigation methods for maize crops in the Hetao Irrigation District, Inner Mongolia Autonomous Region, China. This research delves into the effects of these irrigation methods on carbon dioxide (CO2) exchange and crop growth in this region. The experimental site was divided into drip and flood irrigation zones. The irrigation schedules of this study aligned with the local commonly used irrigation schedule. We employed a developed chamber system to measure the diurnal CO2 exchange of maize plants during various growth stages under both drip and flood irrigation methods. From May to September in 2020 and 2021, two sets of repeated experiments were conducted. In each experiment, a total of nine measurements of CO2 exchange were performed to obtain carbon exchange data at different growth stages of maize crop. During each CO2 exchange measurement event, CO2 flux data were collected every two hours over a day-long period to capture the diurnal variations in CO2 exchange. During each CO2 exchange measurement event, the biological parameters (aboveground biomass and crop growth rate) of maize and environmental parameters (including air humidity, air temperature, precipitation, soil water content, and photosynthetically active radiation) were measured. The results indicated a V-shaped trend in net ecosystem CO2 exchange in daytime, reducing slowly at night, while the net assimilation rate (net primary productivity) exhibited a contrasting trend. Notably, compared with flood irrigation, drip irrigation demonstrated significantly higher average daily soil CO2 emission and greater average daily CO2 absorption by maize plants. Consequently, within the maize ecosystem, drip irrigation appeared more conducive to absorbing atmospheric CO2. Furthermore, drip irrigation demonstrated a faster crop growth rate and increased aboveground biomass compared with flood irrigation. A strong linear relationship existed between leaf area index and light utilization efficiency, irrespective of the irrigation method. Notably, drip irrigation displayed superior light use efficiency compared with flood irrigation. The final yield results corroborated these findings, indicating that drip irrigation yielded higher harvest index and overall yield than flood irrigation. The results of this study provide a basis for the selection of optimal irrigation methods commonly used in the Hetao Irrigation District. This research also serves as a reference for future irrigation studies that consider measurements of both carbon emissions and yield simultaneously.
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-11-13 合作期刊: 《干旱区科学》
摘要： Climate warming has led to the expansion of arable land at high altitudes, but it has also increased the demand for water use efficiency (WUE). To address this issue, the development of water-saving irrigation technology has become crucial in improving water productivity and economic returns. This study aimed to assess the impacts of three irrigation methods on water productivity and economic returns in wolfberry (Lycium barbarum L.) cultivation on the Tibetan Plateau, China during a two-year field trial. Results showed that subsurface irrigation with ceramic emitters (SICE) outperformed surface drip irrigation (DI) and subsurface drip irrigation (SDI) in terms of wolfberry yield. Over the two-year period, the average yield with SICE increased by 8.0% and 2.3% compared with DI and SDI, respectively. This improvement can be attributed to the stable soil moisture and higher temperature accumulation achieved with SICE. Furthermore, SICE exhibited higher WUE, with 14.6% and 4.5% increases compared with DI and SDI, respectively. In addition to the agronomic benefits, SICE also proved advantageous in terms of economic returns. Total average annual input costs of SICE were lower than the other two methods starting from the 8th year. Moreover, the benefit-cost ratio of SICE surpassed the other methods in the 4th year and continued to widen the gap with subsequent year. These findings highlight SICE as an economically viable water-saving irrigation strategy for wolfberry cultivation on the Tibetan Plateau. Thus, this research not only provides an effective water-saving irrigation strategy for wolfberry cultivation but also offers insights into addressing irrigation-related energy challenges in other crop production systems.
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-10-17 合作期刊: 《干旱区科学》
摘要： The unreasonable nitrogen (N) supply and low productivity are the main factors restricting the sustainable development of processing tomatoes. In addition, the mechanism by which the N application strategy affects root growth and nitrate distributions in processing tomatoes remains unclear. In this study, we applied four N application levels to a field (including 0 (N0), 200 (N200), 300 (N300), and 400 (N400) kg/hm2) based on the critical N absorption ratio at each growth stage (planting stage to flowering stage: 22%; fruit setting stage: 24%; red ripening stage: 45%; and maturity stage: 9%). The results indicated that N300 treatment significantly improved the aboveground dry matter (DM), yield, N uptake, and nitrogen use efficiency (NUE), while N400 treatment increased nitrate nitrogen (NO3-N) residue in the 2060 cm soil layer. Temporal variations of total root dry weight (TRDW) and total root length (TRL) showed a single-peak curve. Overall, N300 treatment improved the secondary root parameter of TRDW, while N400 treatment improved the secondary root parameter of TRL. The grey correlation coefficients indicated that root dry weight density (RDWD) in the surface soil (020 cm) had the strongest relationship with yield, whereas root length density (RLD) in the middle soil (2040 cm) had a strong relationship with yield. The path model indicated that N uptake is a crucial factor affecting aboveground DM, TRDW, and yield. The above results indicate that N application levels based on critical N absorption improve the production of processing tomatoes by regulating N uptake and root distribution. Furthermore, the results of this study provide a theoretical basis for precise N management.
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-08-14 合作期刊: 《智慧农业（中英文）》
摘要： ［目的/意义］人工智能（Artificial Intelligence，AI）技术已在学术和工程应用领域掀起了研究高潮，在 地球物理参数和农业气象遥感参数反演方面也表现出了强大的应用潜力。目前大部分AI技术在地学和农学的应用 还是“黑箱”，没有物理意义或缺乏可解释性及通用性。为了促进AI在地学和农学的应用和培养交叉学科的人才， 本研究提出基于AI耦合物理和统计方法的地球物理参数反演范式理论。［方法］首先基于物理能量平衡方程进行 物理逻辑推理，从理论上构造反演方程组，然后基于物理推导构建泛化的统计方法。通过物理模型模拟获得物理 方法的代表性解以及利用多源数据获得统计方法代表性的解作为深度学习的训练和测试数据库，最后利用深度学 习进行优化求解。［结果和讨论］ 判定形成具有通用性和物理可解释的范式条件包括：（1） 输入与输出变量（参 数） 之间必须存在因果关系；（2） 输入和输出变量（参数） 之间理论上可以构建闭合的方程组（未知数个数少于 或等于方程组个数），也就是说输出参数可以被输入参数唯一确定。如果输入参数（变量） 和输出参数（变量） 之 间存在很强的因果关系，则可以直接使用深度学习进行反演。如果输入参数和输出参数之间存在弱相关性，则需 要添加先验知识来提高输出参数的反演精度。此外，本研究以农业气象遥感中的关键参数地表温度、发射率、近 地表空气温度和大气水汽含量联合反演作为案例对理论进行了证明，分析结果表明本理论是可行的，并且可以辅 助优化设计卫星传感器波段组合。［结论］ 本理论和判定条件的提出在地球物理参数反演史上具有里程碑意义。
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-08-14 合作期刊: 《智慧农业（中英文）》
摘要： ［目的/意义］ 农机装备是先进农业生产理念落地的物质支撑，如何提升农机装备设计制造水平及运维管 控能力，充分发挥装备性能，是智慧农业未来发展所面临的核心问题。数字孪生是一种融合多种信息技术、促进 虚实交互融合的先进理念，有助于更加清晰地认识农机装备及其运行过程，从而解决从设计到回收阶段的复杂性 问题，进而全方位地提升农机装备作业质量，更好地满足农业生产需求。［进展］ 首先围绕数字孪生在农机装备领 域的应用，总结数字孪生的研究动态，分析农机装备数字孪生的概念与内涵，提出系统性的体系架构。然后从宏 观发展、系统实现、项目实施多个角度阐述农机装备数字孪生的实现路线。最后介绍农机装备数字孪生的典型应 用场景和案例。［结论/展望］ 数字孪生为农机装备转型升级提供了新方法，为提升农业机械化生产水平提供了新 途径，为实现智慧农业提供了新思路。本文可为农机装备数字孪生相关研究工作的开展提供参考，为数字孪生赋 能智慧农业和智能装备奠定理论基础。
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-08-14 合作期刊: 《智慧农业（中英文）》
摘要： ［目的/意义］ 农业环境动态多变、动植物生长影响因子众多且互作关系复杂，如何将分散无序信息理解 生成生产知识或决策案例是世界性难题。农业知识智能服务技术是应对农业数据低秩化、规则关联度低和推理可 解释性差等现状，提升农业生产全过程综合预测和决策分析能力的核心关键。［进展］ 本文综合分析了感知识别、 知识耦合、推理决策等农业知识智能服务技术，构建由云计算支撑环境、大数据处理框架、知识组织管理工具、 知识服务应用场景组成的农业知识智能服务平台，提出一种基于知识规则和事实案例相结合的农情解析与生产推 理决策方法，构造产前规划、产中管理、收获作业、产后经营等全链条知识智能应用场景。［结论/展望］ 从农业 多尺度农情稀疏特征发现与时空态势识别、农业跨媒体知识图谱构建与自演化更新、复杂成因农情多粒度关联与 多模式协同反演预测、基于生成式人工智能的农业领域大语言模型设计、知识智能服务平台与新范式构建等方面 对农业知识智能服务技术发展趋势进行总结，对实现农业生产由“看天而作”到“知天而作”转变具有技术支撑 作用。
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-08-14 合作期刊: 《智慧农业（中英文）》
摘要： ［目的/意义］为实现不同部位牦牛肉快速、准确识别，本研究提出了一种改进的残差网络模型，并开发 了一种基于智能手机的牦牛肉部位识别软件。［方法］首先对于采集到的牦牛里脊、上脑、腱子、胸肉的原始图像 数据集采用数据增强的方式对其进行扩充，共得到的牦牛肉部位图像17,640张；其次，采用在原网络模型残差块 之后融入轻量级卷积块注意力模块（Convolutional Block Attention Module，CBAM），以加强对不同部位牦牛肉图像 关键细节特征的提取；将原模型最后的全连接层进行改进，以减少后续网络层的连接数，防止出现过拟合，减少 识别图像所需的时间；然后，采用不同的学习率、权重衰减系数和优化器来验证对网络收敛速度和准确率的影响； 最后，开发了移动端App，将改进后的模型部署到移动端。［结果和讨论］通过消融实验，探究出在CBAM、 SENet、NAM、SKNet四种注意力机制模块中，改进效果最好的是CBAM。将改进后的ResNet18_CBAM模型在包含 牦牛里脊、上脑、腱子、胸肉4种不同牦牛肉部位的数据集上进行了试验测试，结果表明，改进后的残差网络模 型在测试集上的识别准确率为96.31%，比改进前的原网络模型提高了2.88%。在手机端的实际场景测试中，牦牛 里脊、上脑、腱子、胸肉的识别准确率分别达到了96.30%、94.92%、98.04%、96.49%。该结果表明，改进后的 ResNet18_CBAM模型可在实际应用中识别不同部位牦牛肉且具有良好的结果。［结论］本研究成果有助于保障牦 牛肉产业的食品质量安全，也为青藏高原地区的牦牛肉产业智能化发展提供技术支撑。
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-08-14 合作期刊: 《智慧农业（中英文）》
摘要： ［目的/意义］目前，对于防风药材产地和品质的鉴别方法主要是根据其物理或化学特征，其方法需对中 药材进行分离提取，存在耗时长，费用高，专业性强，技术难度大等问题，不利于推广应用。随着深度学习的不 断发展，其无需人工提取特征、分类精度高等优点被广泛应用在中药材的识别之中。［方法］针对大多数卷积神经 网络模型在识别防风药材时计算量大、精度低的问题，本研究提出了一种改进的ShuffleNet V2的轻量级防风道地 性识别模型。在不降低网络性能的情况下调整模型架构，减少模型参数量和计算量，用沙漏残差网络（Hourglass Residual Network） 代替传统残差网络，同时引入SE（Squeeze-and-Excitation） 注意力机制，把具有附加信道注意 力的沙漏残差网络嵌入到ShuffleNet V2中，使用SiLU激活函数替换 ReLU 激活函数，丰富局部特征学习，从而提 出轻量化的中药防风道地性识别模型 Shuffle-Hourglass SE。为了验证本文所提出模型的有效性，选用VGG16、 MobileNet V2、ShuffleNet V2和SqueezeNet V2四种经典网络模型进行对比实验。［结果和讨论］结果表明，本研究 提出的模型Shuffle-Hourglass SE获得了最佳性能。在测试集上取得95.32%的准确率、95.28%的召回率，F1分数达 到95.27%，测试时间、模型大小为246.34 ms和3.23 M，不仅在传统CNN网络中是最优的，在轻量级网络中也具 有较大优势。［结论］本研究所提出的模型在保持较高识别精度的同时占用较少的储存空间，有助于在未来的低性 能终端上实现防风道地性的实时诊断。
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-08-14 合作期刊: 《智慧农业（中英文）》
摘要： ［目的/意义］荒漠植物的准确识别是其认识和保护过程中不可或缺的任务，是荒漠生态研究与保护的基 础。自然条件下野外荒漠植物图像的机器视觉自动分类识别可有效提升植物资源调查效率、降低人为主观因素影 响，对荒漠植物的精准分类、多样性保护和资源化利用具有重要意义。［方法］以自然环境下的整株荒漠植物图像 为研究对象，构建新疆干旱区荒漠植物图像数据集，以EfficientNet B0—B4网络为基础网络，提出一种融合迁移学 习和集成学习的荒漠植物图像识别算法，并在公开数据集Oxford Flowers102上进行对比验证。［结果和讨论］基于 EfficientNet B0网络的单一子模型的 Top-1准确率最高可达 93.35%，最低为 92.26%，软投票Ensemble-Soft 模型、硬 投票 Ensemble-Hard 模型以及加权投票法集成的 Ensemble-Weight 模型的准确率分别为 93.63%、93.55%和 93.67%， F1 Score 和准确率相当；基于 EfficientNet B0—B4 网络的单一子模型的 Top-1 准确率最高可达 96.65%，F1 Score 为 96.71%，而 Ensemble-Soft 模型、Ensemble-Hard 模型以及 Ensemble-Weight 模型的准确率分别为 99.07%、98.91%和 99.23%，相较于单一子模型，精度进一步提高，F1 Score 与准确率基本相同，模型性能显著；在公开数据集Oxford Flowers102上进行对比试验，3个集成模型相比 5个子模型准确率和F1 Score 最高提升了 4.56%和 5.05%，最低也提 升了 1.94%和 2.29%，证明了本研究提出的迁移和集成学习策略能够有效提高模型性能。［结论］ 本方法可提高荒 漠植物的识别准确率，通过云端传输至服务器后，实现荒漠植物的准确识别，为真实野外环境下植物图像识别精 度低、模型鲁棒性及泛化性弱等问题提供解决思路。服务于野外调查、教学科普以及科学实验等场景。
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-08-14 合作期刊: 《智慧农业（中英文）》
摘要： ［目的/意义］分析广西甘蔗主产区甘蔗产量与气象因素的关系，利用气象数据预测甘蔗产量，为糖厂及 相关管理部门提供科学的数据支撑。［方法］选用2002~2019年广西五个不同地级市内蔗区的产量数据及14种逐 日气象数据，将每年的各气象因子以78个逐月递增的连续时段的均值与产量进行相关性分析，根据敏感时段分析 法确定关键气象因子，并分析各气象因子在敏感时段对产量的影响。分别利用BP神经网络（BP Neural Network， BPNN）、支持向量机（Support Vector Machine，SVM）、随机森林（Random Forest，RF）、长短期记忆网络（Long Short-Term Memory，LSTM） 建立单蔗区产量预测模型，并采用以全生育期气象均值作为模型输入的方法进行对照 实验。使用HP 滤波法（Hodrick Prescott Filter） 分离出甘蔗气象产量，将5 个蔗区的数据混合，分别利用RF、 SVM、BPNN和LSTM建立通用的多蔗区气象产量预测模型。［结果和讨论］对于单蔗区，敏感时段分析法的模型 预测效果明显优于全生育期取气象均值的方法，LSTM模型对于上述两种数据处理方法的预测效果均明显优于目 前广泛使用的BPNN、SVM、RF模型，敏感时段分析法的LSTM模型整体的均方根误差（Root Mean Square Error， RMSE） 和平均绝对百分比误差（Mean Absolute Percentage Error，MAPE） 分别为10.34 t/ha和6.85%，决定系数Rv 2 为0.8489。对于多蔗区，LSTM预测结果较差，RF、SVM及BPNN三种预测模型都取得了良好的效果，预测效果最 好的BPNN模型的RMSE和MAPE分别为0.98 t/ha和9.59%，Rv 2为0.965。［结论］通过敏感时段分析法筛选的关键 气象因子与产量均呈显著相关，根据敏感时段能准确地分析各气象因子对产量的影响。使用LSTM模型预测单蔗 区产量，使用BPNN模型预测多蔗区甘蔗气象产量的方法是可行的，且预测误差在可接受范围内。
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-08-14 合作期刊: 《智慧农业（中英文）》
摘要： ［目的/意义］烤烟叶片叶绿素含量（Leaf Chlorophyll Content，LCC）是表征烤烟光合作用、营养状况和 长势的重要指标。本研究的目的为高效精确地估测不同生长期烤烟LCC。［方法］以中烟100烟叶为研究对象，利 用无人机搭载Resonon Pika L高光谱成像仪采集烤烟在6个关键生育期冠层反射率数据。基于相关分析筛选了21种 LCC的敏感光谱指数，通过比较不同光谱组合及不同回归分析算法的预测精度，最终建立了基于多种光谱指数组 合的LCC回归估测模型。采用一元线性回归（Unary Linear Regression，ULR）、多元线性回归（Multivariable Linear Regression，MLR）、偏最小二乘回归（Partial Least Squares Regression，PLSR）、支持向量回归（Support Vector Re⁃ gression，SVR）和随机森林回归（Random Forest Regression，RFR）5种建模方法进行LCC估测。［结果和讨论］ 在不同生育期大部分光谱参数与LCC的相关性达到极显著（P<0.01）；相较于传统植被指数，新组合的光谱指数显 著提升了与LCC的相关性；对单变量LCC估测模型ULR，以移栽后75 d新组合的归一化光谱指数与红光比率光谱 指数的单变量建模精度最高，两者决定系数（Coefficient of Determination，R2） 和均方根误差（Root Mean Square Error，RMSE） 分别为0.822和0.814，0.226和0.230。MLR、PLSR、SVR和RFR建模方法预测结果表明，RFR算 法在LCC估测中效果最好，其中使用移栽后75 d数据验证集的R2和RMSE可达0.919和0.146。［结论］本研究通过 分析多种光谱指数与烤烟LCC的响应规律，构建可靠的烤烟叶片LCC估测模型，可为烤烟叶LCC估测以及烤烟的 生长发育监测提供理论依据和技术支撑。
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-08-14 合作期刊: 《智慧农业（中英文）》
摘要： ［目的/意义］快速准确评估作物倒伏灾情状况，需及时获取倒伏发生位置及面积等信息。目前基于无人 机遥感识别作物倒伏缺乏相应的技术标准，不利于规范无人机数据获取流程和提出问题解决方案。本研究旨在探 讨不同空间分辨率无人机遥感影像及特征优化方法对小麦倒伏区域识别精度的影响。［方法］在小麦倒伏后设置3 个飞行高度（30、60和90 m），获取不同空间分辨率（1.05、2.09和3.26 cm） 的数字正射影像图（Digital Ortho⁃ photo Map，DOM） 和数字表面模型（Digital Surface Model，DSM），从不同空间分辨率影像中分别提取5个光谱特 征、2个高度特征、5个植被指数以及40个纹理特征构建全特征集，并选择3种特征选择方法（ReliefF算法、RFRFE 算法、Boruta-Shap算法） 筛选构建特征子集，进而利用3种面向对象监督分类方法——支持向量机（Support Vector Machine，SVM）、随机森林（Random Forest，RF） 和K最近邻（K Nearest Neighbor，KNN） 构建小麦倒伏 分类模型，明确适宜的分类策略，确立倒伏分类技术路径。［结果和讨论］结果表明，SVM的分类效果整体优于 RF和KNN，当影像空间分辨率在1.05~3.26 cm范围内变化时，全特征集和3种优化特征子集均以1.05 cm分辨率 的分类精度最高，优于2.09和3.26 cm。比较发现，Boruta-Shap特征优化方法既能实现降维和提高分类精度的目 标，又能适应空间分辨率的变化，当影像分辨率为3.26 cm 时，总体分类精度相较1.05 和2.09 cm 分别降低了 1.81%和0.75%；当影像分辨率为2.09 cm时，总体分类精度相较1.05 cm降低了1.06%，表现为不同飞行高度下的 分类精度相对差异较小，90 m总体分类精度可达到95.6%，Kappa系数达到0.914，满足了对分类精度的需求。 ［结论］通过选择适宜的特征选择方法，不仅可以兼顾分类精度，还能有效缩小影像空间分辨率变化引起的倒伏分 类差异，有助于提升飞行高度，扩大小麦倒伏监测面积，降低作业成本，为确立作物倒伏信息获取策略及小麦灾 情评估提供参考及支持。
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-08-14 合作期刊: 《智慧农业（中英文）》
摘要： ［目的/意义］针对传统人工识别病虫害存在的效率过低、成本过高等问题，提出一种融合ECA（Effi⁃ cient Channel Attention）注意力机制与DenseNet201的水稻图像识别模型GE-DenseNet（G-ECA DenseNet）。［方法］ 首先在ECA机制上引入Ghost模块的思想构成G-ECA Layer结构，增强其提取特征的能力。其次，在DenseNet201 原有的Dense Block前引入G-ECA Layer，使模型具有更优的通道特征提取能力。由于实验所用的数据集较小，将 DenseNet201在ImageNet数据集上预训练的权重参数迁移到GE-DenseNet中。训练时，采用Focal Loss函数来解决 各分类样本不均衡的问题。同时，使用Adam优化器以避免在模型训练初期由于部分权重随机初始化而导致反向 传播的梯度变化剧烈的问题，在一定程度上削弱了网络训练的不确定性。［结果和讨论］在包含水稻胡麻斑病、水 稻铁甲虫、稻瘟病与健康水稻的3355张图像数据集上进行了实验测试，识别准确率达到83.52%。由GE-DenseNet 模型的消融对比实验可得，引入了Focal Loss函数与G-ECA Layer层之后，模型准确率上升2.27%。将所提模型与 经典NasNet（4@1056）、VGG-16和ResNet50模型相比，分类准确率分别提高了6.53%、4.83%和3.69%；相较于 原始的DenseNet201，对水稻铁甲虫的识别准确率提升达20.32%。［结论］加入G-ECA Layer结构能够使模型更为 准确地捕捉适合于水稻病虫害识别的特征信息，从而使GE-DenseNet模型能够实现对不同水稻病虫害图像更为准 确地识别，为及时防治病虫害，减少各类损失提供技术支持。
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-08-14 合作期刊: 《智慧农业（中英文）》
摘要： ［目的/意义］菠萝的贮藏性与成熟度相关，菠萝采摘前对其成熟度进行识别尤为重要。本研究目的在于 提出一种新型网络模型，提高菠萝成熟度自动识别的准确率和速度。［方法］首先针对菠萝训练数据集样本少与实 时性差等不足，利用在自然环境下拍摄的菠萝照片，自建了种植区场景菠萝成熟度分析数据集。之后将YOLOv4 骨干网络替换成轻量级网络MobileNet V3，提出了轻量级的MobileNet V3-YOLOv4网络。同时训练了原YOLOv4模 型、MobileNet V1-YOLOv4 模型、MobileNet V2-YOLOv4 模型以及Faster R-CNN、YOLOv3、SSD300、Retinanet、 Centernet等五种不同的单、双阶段网络模型，并对比模型的评价指标，分析本文模型的优越性。［结果和讨论］试 验结果表明，MobileNet V3-YOLOv4训练时间为11,924 s，参数量为53.7 MB，训练好的MobileNet V3-YOLOv4在验 证集的平均精度均值（mean Average Precision，mAP） 为90.92%，对于黄熟期菠萝和青熟期菠萝两种类别的检测 精确率（Precision） 分别为100%和98.85%，平均精度（Average Precision，AP） 值分别为87.62%、94.21%，召回 率（Recall） 分别为77.55%、86.00%， F1 分数（F1 Score） 分别为0.87 和0.92， 推理速度（Frames Per Second， FPS）80.85 img/s。［结论］本研究提出的MobileNet V3-YOLOv4实现了在降低训练速度、减小参数量的同时，提高 了菠萝成熟度识别的精度和推理速度，满足实际检测需求。
分类： 农、林、牧、渔 >> 农业基础学科 提交时间： 2023-08-14 合作期刊: 《智慧农业（中英文）》
摘要： ［目的/意义］自然环境中鲜食葡萄的快速识别与精准定位是实现鲜食葡萄机器人自动采摘的先决条件。 ［方法］本研究基于改进的K-means聚类算法和轮廓分析法提出一种鲜食葡萄采摘点自动定位的方法。首先，采用 加权灰度阈值作为聚类算法相似度的判定依据，并以此为基础提出一种自适应调整K值的K-means聚类算法，实 现鲜食葡萄的快速有效识别检测；然后，利用提出的轮廓分析法获得果梗轴和采摘点感兴趣区域，利用几何方法 实现鲜食葡萄采摘点快速准确定位；最后，利用采集的917张鲜食葡萄图像对本研究提出的算法进行实验验证。 ［结果和讨论］本研究提出算法定位的鲜食葡萄采摘点与最优采摘点的误差小于12个像素的成功率为90.51%，平 均定位时间为0.87 s，实现鲜食葡萄采摘点的快速准确的定位。在篱壁式种植方式与棚架式种植方式下分别进行 50次模拟仿真试验，结果表明，篱壁式紫葡萄采摘点定位成功率为86.00%，棚架式紫葡萄识别定位成功率达到 92.00%，篱壁式绿葡萄采摘点定位成功率为78.00%，棚架式绿葡萄识别定位成功率为80.00%，整体试验效果较 好。［结论］本研究可为鲜食葡萄采摘机器人实现精准采摘葡萄提供技术支撑。
点击量
通过
点击量
下载量
评论
