Subjects: Geosciences >> Geography submitted time 2022-03-28 Cooperative journals: 《干旱区研究》
Abstract：地貌形态对象往往大小悬殊，跨越特定的空间尺度。现有的地貌形态自动分类方法尚未充分顾及该特点，分类精度受到制约。利用地貌形态的大小为尺度，提出1种涉及尺度跨越性的地貌形态多尺度综合分类方法，该方法由多尺度分割、按尺度顺序筛选和多尺度合并3个步骤构成。其中，按尺度顺序筛选是1个以多尺度特征提取和监督分类为基础、以小尺度（小尺寸）优先和概率最大化为准则的被分类对象迭代确认过程。以黄土高原为例的试验结果表明，该方法简便可靠（总体精度可以达到75.16%，Kappa系数可以达到0.71），可用于地貌形态精细化分类。
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-12-13 Cooperative journals: 《计算机应用研究》
Abstract： Aiming at the single description and the noise sensitive problems of the local binary pattern(LBP) , a multi-scale adaptive threshold local ternary pattern(MSALTP) algorithm is proposed. The algorithm first enlarges the original images . Secondly, divides the images into several regions equally and calculates the mean value of the pixels. Then calculates the deviation between the center and the mean pixel value of each region. Finally, extract the ALTP features and the resulting statistical features histograms are used to classify the images. Experiments show that the proposed algorithm recognition rates are higher than the current anti-noise algorithms under different noise.
Subjects: Astronomy >> Astrophysical processes submitted time 2018-05-15 Cooperative journals: 《天文研究与技术》
Abstract：中国新一代频电频谱日像仪-明安图射电频谱日像仪（Mingantu Spectral Radioheliograph，MUSER）以高时间、高空间、高频率分辨率工作在0.4GHz~15GHz，为太阳爆发活动初始能量释放区的物理过程、太阳电子加速等研究开辟了新的窗口。高性能高质量太阳成像算法是MUSER数据处理流水线至关重要的研究内容。参考法国墨东天文台太阳干涉阵的数据处理方法，系统讨论分析了多尺度（Multi-Scale）CLEAN算法，给出了适用于MUSER的多尺度CLEAN算法参数，并重点讨论了算法的GPU并行。实验结果表明，改进的多尺度CLEAN在算法效率上比基于GPU实现的Högbom CLEAN提高了近3倍，有效提高了整个MUSER数据处理流水线的性能。
Subjects: Mechanics >> Applied Mechanics submitted time 2023-03-20 Cooperative journals: 《应用力学学报》
Abstract： Rock deformation and failure occur with the increase of extemal load. When the external loadremains unchanged, the deformation does not stop, and it will continue to increase , resulting in creep de-formation.'The fundamental cause is the heterogeneity of rock.Rock creep fracture is a process of micro-scale fracture , meso-scale crack propagation and penctration , macro-scale deformation increase and failureprocess caused by the original heterogeneity of rock.'Therefore , it is necessary to study the multi-scale evo-lution mechanism of rock creep fracture.'Taking sandstone as the research object , this paper analyzes thedeformation and fracture mechanism of sandstone samples at different scales from the macro-evolution ex-periment of creep fracture and the micro-scanning experiment of fracture.Through the macroscopic evolution experiment of sandstone creep fracture ,it can be seen that there are no cracks on the surface of rockin the initial creep and constant velocity creep stages , but mainly the initiation and propagation of micro-and micro-cracks.When the macro-fracture surface is formed , it enters the accelerated creep stage. In theaccelerated creep stage , friction sliding occurs along the fracture surface ,and finally failure occurs. The mi-croscopic analysis of the micro-composition and fracture characteristics of the sandstone can be obtained byanalyzing the micro-composition and micro-fracture characteristics of the sandstone.The results of multi-scale analysis directly reflect the spatial position and direction of crack propagation in rock. This is of greatsignificance for further study of the evolution mechanism of rock creep rupture.
Subjects: Geosciences >> Geography submitted time 2023-08-01 Cooperative journals: 《干旱区地理》
Abstract：随着互联网技术的高速发展，城市间信息联系更加紧密，空间结构不断重构。本文以2012年与2019年黄河流域内部城市之间及与外部城市相互搜索的百度指数为数据源，采用社会网络分析及空间分析等方法，探讨了本地与全国2个尺度下黄河流域城市网络的时空演化格局。结果表明：（1） 全国尺度下的城市点度中心性与本地尺度高度相关，但相较于本地尺度，中心性高值区愈发向东部集聚，空间非均衡性增加，西安市、青岛市核心地位更加突出。（2） 全国尺度下的城市中介中心性与本地尺度不相关，且相较于本地尺度，中心性高值区由西向东转移，空间非均衡性降低，西安市中介效应优势更加突出。（3） 黄河流域内部联系呈现以西安市-郑州市及青岛市-济南市为核心的相关孤立的骨干网络，域外高层级联系愈发向长三角、成渝等南方城市群集中。（4） 相较于本地尺度，全国尺度下黄河流域首位联系空间组织变化不大，第二位联系大多从流域中心或省域次中心转向链接北京市，第三位联系则大多从邻近链接转向择优链接域外中心城市。研究结果可为协同推进黄河流域高质量发展提供决策参考。
Subjects: Mechanics >> Basic Mechanics submitted time 2023-04-24 Cooperative journals: 《应用力学学报》
Abstract： Real machined surfaces are always rough across a wide range of length scales.This paper develops a hierarchical contact model,in which surface power spectrum density function is employed to divide roughness into two parts.In the macroscopic part,by cutting off short-wavelength roughness,the approximate contact area and the distribution of normal force are obtained adopting the continuum mechanics solution.In the microscopic part,the superposition of asperities for short wavelength is transformed into a two dimensional Cantor set based on fractal theory,and the approximate contact area is further refined based on the local pressure and geometric parameters.The contact response of the whole surface is thus obtained by adding the modification of the microscopic part into the macroscopic part.For verification,FEM is adopted to calculate the response of the fractal rough surfaces.The results demonstrate that our model can adequately predict the contact response of the multi-scale rough surface.The current study reduces the computing scale of the mechanical model of rough surfaces and thus provides a new method modeling rough surfaces in a larger scale span.This may have significant implications for further research on contact and friction between multi-scale rough surfaces.
Subjects: Survey & Drawing Science and Technology >> Photogrammetry and Remote Sensing Subjects: Medicine, Pharmacy >> Preventive Medicine and Hygienics Subjects: Statistics >> Biomedical Statistics submitted time 2020-02-19
Abstract： " "
Peer Review Status:
Awaiting Review
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-17 Cooperative journals: 《计算机应用研究》
Abstract： The research of multi-scale data mining mainly applied to space remote sensing image data sets, and conduct scale division based on the resolution or regional segmentation of the images, then analysis knowledge on each scale layer. Recently, there are quite a few learners applied the multi-scale data mining to general data sets, and conduct scale division based on the level theory, concept hierarchy and inclusion degree etc. , study the distribution rule on different scale layers, and then found significant facts. For example, multi-scale association rules, multi-scale clustering. But it has not been involved in the field of the classification mining. This paper defines the concept of generalized fractal interpolation theory, break the situation that limited to the use of the iteration function system(IFS) , and extend the application of the fractal interpolation. Then, a multi-scale classification scaling-down algorithm based on the generalized fractal interpolation theory named MSCSDA (Multi-Scale Classification Scaling-Down Algorithm) is proposed. This paper performs experiments on four UCI benchmark data sets, and one real data set (H province part of the population) . Then analysis the experimental results compare MSCSDA with KNN, Decision Tree and LIBSVM algorithms on different data sets. The experimental results show that the MSCSDA algorithm gives better results in terms of classification than the others.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-24 Cooperative journals: 《计算机应用研究》
Abstract： In order to overcome the difficulties of detecting the Uyghur text in natural scene images, this paper improved a single deep neural network to detect Uyghur text in natural scene images. The network structure combined the Uyghur feature extraction and the multi-layer features fusion text detection component. What was more, it predicted Uyghur text bounding box and the confidence score of Uyghur text in an end-to-end manner. Uyghur feature extraction component used convolutional neural network to extract multi-scale and multi-level Uyghur features from natural Uyghur images. The multi-layer features fusion text detection component made use of the features extracted by the Uyghur feature extraction component to predict the position of the Uyghur text bounding boxes and the confidence of the Uighur category. The analysis shows that Uyghur text had more special features than English and Chinese texts. For this feature, it designed a default box with multiple aspect ratios and adjusted multiple sizes and the size of some convolution kernels. Experiments on Uyghur natural scene images collected by pattern recognition and Intelligent control laboratory of electrical engineering college of Xinjiang university show that the improved single deep neural network method considers the influence of multi-scale and multi-level images on the detection accuracy and improves the detection accuracy. The accuracy and the F value of the algorithm respectively reach 0.723 4 and 0.611 5.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-17 Cooperative journals: 《计算机应用研究》
Abstract： Retinal vessel segmentation is the basis of ophthalmic computer-aided diagnosis and large-scale ophthalmic disease screening systems. In order to assist ophthalmologists in the diagnosis of fundus diseases, this paper presented a retinal vessel segmentation method based on PST and multi-scale Gaussian filter. Firstly, it enhanced the green channel of color fundus image; then, it used Gaussian filters with different scales to reduce the noise of retinal vessels with different diameters after preconditioning, and combined with the Gaussian filters, it used the edge detection algorithm of the PST(phase stretch transform) to obtain preliminary retinal vascular segmentation map; finally, it integrated the initial retinal vessel segmentation maps and denoised the result based on morphology to obtain the final retinal vessel segmentation map. The experiments on DRIVE retinal image database show that the average accuracy is 93%, the average sensitivity is 77%, and the average specificity is 95%. The experimental results verify the effectiveness of the method proposed.
Subjects: Mechanics >> Other Disciplines of Mechanics submitted time 2024-06-17 Cooperative journals: 《应用力学学报》
Abstract：
The combined resonance of superharmonic resonance and 1:3 internal resonance of a rectangular thin plate in transverse magnetic field is studied.For the rectangular thin plate with one side fixed and three sides simply supported,the two degree of freedom nonlinear vibration differential equations are obtained by the Galerkin method.Then the multiple-scale method is employed to solve equations,and amplitude frequency response equations of the first two modes are obtained.Through calculation examples,the relationship between the first two order response amplitudes and parameters when superharmonic internal resonance occurs in system is obtained,in which the influence of parameters e.g.,external excitation amplitude and magnetic field intensity on the vibration of system is discussed.The results show that the high-order modes are indirectly excited when internal resonance is considered,and there is energy exchange in the system,and the resonance of system can be controlled by changing the magnetic field intensity.
Subjects: Materials Science >> Materials Science (General) submitted time 2023-03-19 Cooperative journals: 《金属学报》
Abstract： To solve such problems as sensitivity to noise and low accuracy of grain size evaluation using traditional ultrasonic time-domain attenuation method, an ultrasonic nondestructive evaluation model based on multiscale attenuation coefficient was proposed. The distribution of time- scale of ultrasonic energy was obtained by means of wavelet transformation, then to calculate the distribution of attenuation coefficient with scale, and to make a comprehensive analysis of attenuation characteristics of various scales. After the weighted multi-scale ultrasonic attenuation coefficient was defined, a multi-scale ultrasonic attenuation evaluation model was established on the basis of combination of optimal dimension and normalized weight distribution strategy designed by particle swarm optimization. 304 stainless steel was used in the test. The distribution of attenuation coefficient with scale shows that ultrasonic wave of small scales attenuates fast, presenting the frequency characteristics of ultrasonic attenuation among high scattering materials. Following increase of the sample grain size, ultrasonic attenuation of all scales was intensified significantly. Test results show that the sound velocity method, the traditional evaluation method and the proposed method have maximum systematic errors of +12.57%, +5.85% and -1.33%, respectively. With these 3 methods, evaluation results of the sample with a mean grain size of 103.5 mm measured by metallographic method are (110.4�7.8), (98.2�6.6) and (101.7�3.9) mm, respectively, showing that the presented method can not only reduce the systemic error, but also can effectively control the random error by constant Q filtering properties of wavelet transformation. This model can be extended to grain size evaluation of other metals.
Subjects: Materials Science >> Materials Science (General) submitted time 2022-06-20
Li, Xiangyan
Li, Xiaolin
Zhang, Yange
Xu, Yichun
Wu, Xuebang
Xie, Zhuoming
Wang, Xianping
Pan, Bicai
Liu, Changsong
Fang, Qianfeng
Chen, Junling
Luo, Guangnan
Wang, Zhiguang
Abstract：
The behavior of nano-voids composed of vacancies (Vs) at grain boundaries (GBs) is fundamental to the design of the radiation tolerance of poly-crystalline metals (PCs) via GB engineering. In this study, based on differential evolution, a framework for determining the stable structure of GB nano-voids is developed. Combining the framework with multiscale simulations, we elucidate the vacancy-accumulation and GB void
formation mechanism under irradiation. A GB-structure dependent picture is revealed. At special coincidence-site-lattice (CSL) GBs of Ʃ5(310) and Ʃ5(210) with a medium V-GB binding energy, the V could be reemitted from the GB and also has driving force to be clustered at the GB, developing particularly stable V-clusters from a linear configuration to a platelet and finally to three-dimensional void that has large strain fields in iron with small bulk modulus and a bulk-void alike structure in the GB with large bulk modulus. A group of vacancies reconstruct their positions during the growth. The ripening is also mediated by the mobility of small V-clusters in addition to free Vs. General high-angle and low-angle GBs trap Vs efficiently, where V-clusters only align one-dimensionally or hardly nucleate. Based on the bonding among the vacancies and their neighboring atoms of a nano-void, we propose a high-accuracy predictive linear energetic model applied to the nano-void both at the iron/molybdenum/tungsten GBs and in the grain interior. The model captures the anisotropic feature of a nano-void and reproduces the oscillated vacancy energy level near a nano-void, showing distinct advantages over conventional continuum model and Wulff construction based energy model. Finally, the collective behavior of multiple GBs plays a role in the GB void formation. The present work offers fundamental mechanistic insights to GB nano-void formation and growth and sets a key step towards GB-void prevention in PCs by reducing the fraction of special CSL-GBs.
Peer Review Status:Awaiting Review
Subjects: Geosciences >> Other Disciplines of Geosciences submitted time 2019-09-11 Cooperative journals: 《干旱区地理》
Abstract：为研究沙地信息提取的方法，采用基于CART决策树的面向对象方法，提取中卫市沙坡头区的沙地信息。首先对研究区进行多尺度分割和光谱差异分割得到对象层，然后选择合适的提取特征和训练样本点，最后输入选择的提取特征和样本点生成CART规则树，并对地物进行分类，提取出沙地信息。结果表明：采用面向对象的CART决策树方法提取沙地信息具有较高自动化程度和精确度，依此构建的CART决策树总体分类精度可达到77%，是最近邻分类结果的1.12倍，支持向量机分类结果的1.57倍，此外，NDBI（归一化裸露指数）、GSI（粒度指数）和SWIR 2（第七波段）均值可以成功的将沙地、戈壁和裸岩石砾地三个易混地物区分开来，是沙地提取过程中三个重要的特征指数。
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-24 Cooperative journals: 《计算机应用研究》
Abstract： To solve the problem that lung nodule automatic detection methods for CT images can only get low sensitivities with a lot of false positives, this paper proposed an integrated method with hybrid loss based 3D fully convolutional network as candidate detection and attention-based multi-scale residual network as nodule classification. Firstly, this paper established a 3D fully convolutional network based on dice coefficient loss, and the network filtered hard negative samples uniting with positives to fine-tune. Then, it designed an attention based multi-scale 3D residual convolutional network to classify the candidate and recognize true nodules. Experiment results on the LUNA16 dataset show that the proposed method achieves the sensitivity of 97.18% at 59.1 false positives per scan in the candidate detection stage, and the whole system achieves the average sensitivity of 0.880, which demonstrates our proposed method can improve sensitivity with fewer false positives and achieve superior performance.
Subjects: Statistics >> Social Statistics submitted time 2023-12-04 Cooperative journals: 《智慧农业（中英文）》
Abstract： Objective The root system is an important component of plant composition, and its growth and development are crucial for plants. Root image segmentation is an important method for obtaining root phenotype information and analyzing root growth patterns. Research on root image segmentation still faces difficulties, because of the noise and image quality limitations, the intricate and diverse soil environment, and the ineffectiveness of conventional techniques. This paper proposed a multi-scale feature extraction root segmentation algorithm that combined data augmentation and transfer learning to enhance the generalization and universality of the root image segmentation models in order to increase the speed, accuracy, and resilience of root image segmentation. Methods Firstly, the experimental datasets were divided into a single dataset and a mixed dataset. The single dataset acquisition was obtained from the experimental station of Hebei Agricultural University in Baoding city. Additionally, a self-made RhizoPot device was used to collect images with a resolution pixels of 10,200×14,039, resulting in a total of 600 images. In this experiment, 100 sheets were randomly selected to be manually labeled using Adobe Photoshop CC2020 and segmented into resolution pixels of 768× 768, and divided into training, validation, and test sets according to 7:2:1. To increase the number of experimental samples, an open source multi-crop mixed dataset was obtained in the network as a supplement, and it was reclassified into training, validation, and testing sets. The model was trained using the data augmentation strategy, which involved performing data augmentation operations at a set probability of 0.3 during the image reading phase, and each method did not affect the other. When the probability was less than 0.3, changes would be made to the image. Specific data augmentation methods included changing image attributes, randomly cropping, rotating, and flipping those images. The UNet structure was improved by designing eight different multi-scale image feature extraction modules. The module structure mainly included two aspects: Image convolution and feature fusion. The convolution improvement included convolutional block attention module (CBAM), depthwise separable convolution (DP Conv), and convolution (Conv). In terms of feature fusion methods, improvements could be divided into concatenation and addition. Subsequently, ablation tests were conducted based on a single dataset, data augmentation, and random loading of model weights, and the optimal multi-scale feature extraction module was selected and compared with the original UNet. Similarly, a single dataset, data augmentation, and random loading of model weights were used to compare and validate the advantages of the improved model with the PSPNet, SegNet, and DeeplabV3Plus algorithms. The improved model used pre-trained weights from a single dataset to load and train the model based on mixed datasets and data augmentation, further improving the model's generalization ability and root segmentation ability. Results and Discussions The results of the ablation tests indicated that Conv_ 2+Add was the best improved algorithm. Compared to the original UNet, the mIoU, mRecall, and root F1 values of the model increased by 0.37%, 0.99%, and 0.56%, respectively. And, comparative experiments indicate Unet+Conv_2+Add model was superior to the PSPNet, SegNet, and DeeplabV3Plus models, with the best evaluation results. And the values of mIoU, mRecall, and the harmonic average of root F1 were 81.62%, 86.90%, and 77.97%, respectively. The actual segmented images obtained by the improved model were more finely processed at the root boundary compared to other models. However, for roots with deep color and low contrast with soil particles, the improved model could only achieve root recognition and the recognition was sparse, sacrificing a certain amount of information extraction ability. This study used the root phenotype evaluation software Rhizovision to analyze the root images of the Unet+Conv_2+Add improved model, PSPNet, SegNet, and DeeplabV3Plu, respectively, to obtain the values of the four root phenotypes (total root length, average diameter, surface area, and capacity), and the results showed that the average diameter and surface area indicator values of the improved model, Unet+Conv_2+ Add had the smallest differences from the manually labeled indicator values and the SegNet indicator values for the two indicators. Total root length and volume were the closest to those of the manual labeling. The results of transfer learning experiments proved that compared with ordinary training, the transfer training of the improved model UNet+Conv_2+Add increased the IoU value of the root system by 1.25%. The Recall value of the root system was increased by 1.79%, and the harmonic average value of F1 was increased by 0.92%. Moreover, the overall convergence speed of the model was fast. Compared with regular training, the transfer training of the original UNet improved the root IoU by 0.29%, the root Recall by 0.83%, and the root F1 value by 0.21%, which indirectly confirmed the effectiveness of transfer learning. Conclusions The multi-scale feature extraction strategy proposed in this study can accurately and efficiently segment roots, and further improve the model's generalization ability using transfer learning methods, providing an important research foundation for crop root phenotype research.
Subjects: Other Disciplines >> Synthetic discipline submitted time 2023-03-28 Cooperative journals: 《中国科学院院刊》
Abstract： As China’s environmental protection emission standards are becoming stricter and industrial parks are taking shape, problems such as the lack of stable cooperative treatment technology for toxic pollutants and carbon emission reduction, and the high control cost seriously restrict the sustainable development of economic society and the realization of the strategic goal of carbon emission reduction. Guided by the major environmental protection demand of industries, this study put forward the method, strategy and scientific basis of “Synergistic reduce pollution and carbon in the whole process of industry”. Through the coordination of control methods, cross-media, multi-field and multi-factor modeling optimization, the creative discovery of basic science at molecular level was directly linked with engineering research, which provided new scientific support for the synergy reduction of pollution and carbon emission, and contributed theoretical methods to China’s industrial green development.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-18 Cooperative journals: 《计算机应用研究》
Abstract： Hashing has been widely used for large-scale multimedia retrieval because of its advantages of storage and retrieval efficiency. The use of the semantic similarity improving the hash coding quality has recently been more widely concerned. Traditional supervised hash methods for image retrieval represent an image as a manual feature vector or a machine learning feature vector, and then perform a separate quantization step to generate a binary code. Such methods do not control the quantization error effectively, and cannot guarantee the balance of hash code. To this end, this paper presents a new multi-scale balanced deep hash method. The method uses multi-scale input, which effectively improves the ability of learning the image features from the network. Moreover, a new loss function is proposed. Under the premise of preserving the semantic similarity, the quantization error and the balance of hash code are taken into account to generate the high quality hash code. After experimenting on two benchmark databases: CIFAR-10 and Flickr, this method has been improved by 5.5% and 3.1% of the search accuracy compared with today's advanced image retrieval methods.
Subjects: Traffic and Transportation Engineering >> Ship Engineering submitted time 2024-03-23
Abstract： In recent years, there has been an increasing demand for higher detection and classification accuracy of ship targets to enable safe ship navigation, driving the development of ship intelligence. However, the performance of deep learning-based ship target detection algorithms is affected by the optical imaging process of ship targets, which can be easily disrupted by environmental factors such as wind, current, rain, and fog. Additionally, the diverse range of ship types, morphologies, and sizes pose challenges for accurate detection and identification of ship targets. To address these challenges, this paper proposes a multi-scale neural network-based target detection method for improving the accuracy of ship target detection in optical images. The proposed method employs a Convolutional Neural Networks (CNN) to extract image features. The improved backbone of CSPDarkNet and multi-scale network is used to realize the accurate detection of the ship-borne optical camera on the water ship target, and the detection accuracy of the model for small targets and dense targets is improved. Furthermore, label smoothing to prevent overfitting, and non-maximum suppression to reduce repetitive detections. Experimental results demonstrate that the proposed model achieves accurate detection of ship targets on water and can be used for the detection of small and intensive targets. The mean average precision (mAP) of the proposed method on the Ship-Detection dataset reaches 84.80, which outperforms previous research methods such as Faster-RCNN, DINO and offers greater potential for practical applications.
Peer Review Status:Awaiting Review
Subjects: Agriculture, Forestry,Livestock & Aquatic Products Science >> Other Disciplines of Agriculture, Forestry,Livestock & Aquatic Products Science submitted time 2024-06-17 Cooperative journals: 《智慧农业（中英文）》
Abstract： Objective Currently, the lack of computerized systems to monitor the quality of cabbage transplants is a notable shortcoming in the agricultural industry, where transplanting operations play a crucial role in determining the overall yield and quality of the crop. To address this problem, a lightweight and efficient algorithm was developed to monitor the status of cabbage transplants in a natural environment. Methods First, the cabbage image dataset was established, the cabbage images in the natural environment were collected, the collected image data were filtered and the transplanting status of the cabbage was set as normal seedling (upright and intact seedling), buried seedling (whose stems and leaves were buried by the soil) and exposed seedling (whose roots were exposed), and the dataset was manually categorized and labelled using a graphical image annotation tool (LabelImg) so that corresponding XML files could be generated. And the dataset was pre-processed with data enhancement methods such as flipping, cropping, blurring and random brightness mode to eliminate the scale and position differences between the cabbages in the test and training sets and to improve the imbalance of the data. Then, a cabbage transplantation state detection model based on YOLOv8s (You Only Look Once Version 8s) was designed. To address the problem that light and soil have a large influence on the identification of the transplantation state of cabbage in the natural environment, a multi-scale attention mechanism was embedded to increase the number of features in the model, and a multi-scale attention mechanism was embedded to increase the number of features in the model. Embedding the multi-scale attention mechanism to increase the algorithm’s attention to the target region and improve the network’s attention to target features at different scales, so as to improve the model’s detection efficiency and target recognition accuracy, and reduce the leakage rate; by combining with deformable convolution, more useful target information was captured to improve the model’s target recognition and convergence effect, and the model complexity increased by C3-layer convolution was reduced, which further reduced the model complexity. Due to the unsatisfactory localization effect of the algorithm, the focal extended intersection over union loss (Focal-EIoU Loss) was introduced to solve the problem of violent oscillation of the loss value caused by low-quality samples, and the influence weight of high-quality samples on the loss value was increased while the influence of low-quality samples was suppressed, so as to improve the convergence speed and localization accuracy of the algorithm. Results and Discussions Eventually, the algorithm was put through a stringent testing phase, yielding a remarkable recognition accuracy of 96.2% for the task of cabbage transplantation state. This was an improvement of 2.8% over the widely used YOLOv8s. Moreover, when benchmarked against other prominent target detection models, the algorithm emerged as a clear winner. It showcased a notable enhancement of 3% and 8.9% in detection performance compared to YOLOv3-tiny. Simultaneously, it also managed to achieve a 3.7% increase in the recall rate, a metric that measured the efficiency of the algorithm in identifying actual targets among false positives. On a comparative note, the algorithm outperformed YOLOv5 in terms of recall rate by 1.1%, 2% and 1.5%, respectively. When pitted against the robust faster region-based convolutional neural network (Faster R-CNN), the algorithm demonstrated a significant boost in recall rate by 20.8% and 11.4%, resulting in an overall improvement of 13%. A similar trend was observed when the algorithm was compared to the single shot multibox detector (SSD) model, with a notable 9.4% and 6.1% improvement in recall rate. The final experimental results show that when the enhanced model was compared with YOLOv7-tiny, the recognition accuracy was increased by 3%, and the recall rate was increased by 3.5%. These impressive results validated the superiority of the algorithm in terms of accuracy and localization ability within the target area. The algorithm effectively eliminates interferenced factors such as soil and background impurities, thereby enhancing its performance and making it an ideal choice for tasks such as cabbage transplantation state recognition. Conclusions The experimental results show that the proposed cabbage transplantation state detection method can meet the accuracy and real-time requirements for the identification of cabbage transplantation state, and the detection accuracy and localization accuracy of the improved model perform better when the target is smaller and there are weeds and other interferences in the background. Therefore, the method proposed in this study can improve the efficiency of cabbage transplantation quality measurement, reduce the time and labor, and improve the automation of field transplantation quality survey.
Hits
Hits
Downloads
Comment
YES