Efficient and accurate identification and position of compound buds can greatly contribute to the orchard robots for automatic bud removal operations, thereby improving the efficiency of flower and fruit thinning. This study aims to accurately identify and position the small objects in complex backgrounds for grape bud removal in the field. (1) A visual recognition system was proposed for the concurrent bud and secondary bud using a combination of far and near visual. Euclidean distance was applied on the visual mapping plane. The concurrent bud was then identified in the far view, whereas, the secondary bud was identified in the near view within the concurrent bud mapping area, in order to improve the recognition accuracy of small target objects in the field. The original images (2132 images) were randomly divided into the training set (1735 images) and validation set (397 images) in the ratio of 8:2. The training set and validation set were uniformly cropped to 2944×1656 pixel size, and then scaled to 1280×720 pixel size with 2.3 times of equal proportion. Data augmentation was carried out by three random combinations of augmentation through scaling, flipping, and color gamut transformation for the concurrent and secondary bud datasets. Once the augmentation scale was too large for the target features, the images were manually excluded from the dataset. As such, the training set was enlarged to 15840 images. YOLOv5m and YOLOv5s were selected as the network models for juxtaposed bud and parabudding detection. The sizes of the trained models were 42.1 and 14.4 MB, respectively, with AP of 0.702 and 0.773, F1 scores of 0.685 and 0.765 on the test set of concurrent and secondary bud images, respectively, where the average inference time per image was 10.62 and 7.01 ms. 60 compound bud images showed that the overall average accuracy of this improved model was 0.905, and the average detection time was 18.1 ms for each process; (2) An online detection and position were then proposed for the compound bud in the field. The online detection and position were introduced in the prediction stage to improve the light-resistant ability of the depth camera when positioning in a complex environment, in order to reduce the fluctuation and error of the depth information. The same compound bud was online detected and positioned for 5 consecutive frames. The initial frame and the latest frame image data were deleted and updated in real time, until the relative error of positioning coordinates in 5 consecutive frames was less than the error threshold, in order to obtain the average positioning coordinates of the compound bud. Better performance was achieved in the positioning accuracy of compound bud in natural background and the resistance to light coupling in complex light environment. The test results showed that the positioning accuracy of the improved model was ±0.916 and ±0.654 mm for the concurrent and secondary bud, respectively. The repeatability and accuracy can fully meet the demand for the secondary bud wiping, indicating a reliable and effective system. The finding can provide high-quality technical support for the online positioning of the concurrent and secondary buds for sprout wiping operation in sprouting robots, in order to realize the automatic sprout wiping in orchards.

Cropland parcels are the basic unit for agricultural production, and their size and shape may change due to human activities, e.g., land consolidation. Remote sensing has been increasingly used for mapping cropland parcel, yet detecting changes in cropland parcels by wall-to-wall mapping is time-consuming. This paper proposes a new algorithm to identify whether and where cropland parcel changes have been undertaken without generating complete parcel maps. We use the number of edge pixels derived from remote sensing imagery as a proxy indicator for cropland parcel changes. First, we apply a Sobel operator to delineate the total edge pixels of parcels from dual-time images. Second, we apply the connected-components labeling to remove pseudo-edges arising from non-cropland built structures and transmission towers. We then perform topological optimization, including morphological dilation and skeleton extraction, to eliminate redundant edge pixels for parcel structure. Finally, we detect whether parcel changes have been undertaken by counting and comparing the number of edge pixels derived from dual-time images. We applied this innovative framework in five regions in East Asia where land consolidation has significantly changed cropland parcels. Our method demonstrated robust detection results, with stable accuracy, precision, recall, and F1-score, all exceeding 0.85. Screening redundant edge pixels reduced noise and permitted efficient detection of changes in cropland parcels. Our method extends the traditional detection of semantic change to structural change and can quickly detect cropland parcel changes with high accuracy. This capability offers the potential to identify hotspot areas of cropland changes on a larger scale without the need to produce full cropland maps, which is particularly useful for monitoring land consolidation programs.

Stripe rust has posed a serious threat to the wheat yield in recent years. It is crucial to breed the wheat varieties resistant to stripe rust. However, the identification of resistance is single and inefficient in traditional breeding. In this study, an efficient identification was proposed to determine the different resistance grades to the stripe rust using the time series of vegetation index response to wheat canopy. An unmanned aerial vehicle (UAV) was utilized to collect multi-temporal spectral images of the canopy in the naturally occurring breeding populations of colony wheat (600 samples in total, 516 genotypes). Six sensitive features were screened for the severity of stripe rust disease using Random Forest and ReliefF algorithms: normalized pigment chlorophyll index (NPCI), woebbecke index (WI), chlorophyll index rededge (CIrededge), (green atmospherically resistant index GARI), normalized difference vi (NDVI), and chlorophyll index green (CIgreen). These indices were verified as sensitive features. The severity of stripe rust disease incidence was dynamically characterized using the time series of these indices in the test population. The support vector machine (SVM) was used to establish a classification model for the severity grade of stripe rust disease, according to the sensitive features. There was a very small difference in the performance of the test set and the unscreened original features, indicating the effectiveness of the screened sensitive features. The time series of six sensitive traits was observed in the samples of different resistance grades. It was found that there were no significant differences in the CIgreen and CIrededge among the samples with the different resistance grades. This indicated that the saamples were not applicable to categorize the resistance grades to stripe rust. The differences exhibited by GARI, NDVI, NPCI and WI were used to classify the resistance grades to stripe rust. General machine learning cannot capture the smaller differences of feature variation in the samples with the different resistance grades. Therefore, an improved mode was proposed to extract the features from two-dimensional images that transformed vegetation index time series, in order to realize the classification of stripe rust resistance grade. Four time-series vegetation indices (NPCI, GARI, NDVI, and WI) were better distinguished the different disease resistance grades among the sensitive features, and then used to generate the Gramian Angular Summation Field (GASF) images by the Gramian Angular Field. Data augmentation was performed on the dataset to equalize the number of samples in each resistance grade. Each dataset had a total of 1 040 samples, and was then divided into four grades of stripe rust resistance, where each grade contained 260 sample images, while each dataset was divided into training, validation, and testing sets in the ratio of 6:2:2. DenseNet121 model was separately trained using each dataset, in order to classify the various stripe rust resistance. A better performance was achieved in the classification models with the GASF_NPCI and GASF_WI as the input features, compared with the GASF_GARI and GASF_NDVI. The model with the GASF_NPCI as a feature was slightly less effective in distinguishing the samples with the resistance grades R and MR, where the precision and recall were relatively low. There was no difference in the models with the GASF_WI for the precision and recall of the samples that predicted each stripe rust resistance grade. In the F1 scores of the test set, the different vegetation indices on the resistance grades of stripe rust in colony wheat were ranked in the order of NPCI, WI, GARI, NDVI. The classification model with the GASF_NPCI was the most effective in the test set, with an F1 score of up to 0.833. There was a better distinction of differences in the stripe rust resistance grades among different varieties (lines) of population wheat. The grades of wheat stripe rust resistance were fully identified using time series of spectral vegetation index. Meanwhile, the finding can also provide a strong reference for the disease resistance breeding of crops.

Wine grape has been one of the most significant economic crops. The precise identification of wine grape varieties can greatly contribute to effective vineyard management and the quality of the wine industry. It is still lacking in the adaptability to new varieties in the variety identification, particularly for the labeled data and the substantial costs. In this study, a two-phase variety identification was proposed using few-sample learning, including the extraction phase and meta-learning phase. Firstly, the impact of complex backgrounds was mitigated by the few-sample learning model. A Deeplabv3+ semantic segmentation model was then developed on the segmented image after post-processing. Specifically, image cropping was performed to enlarge the pixel area of the leaves in the image. The precise extraction of the foreground leaves was provided on the high-quality image inputs for the subsequent model. Secondly, metric meta-learning was utilized to evaluate the similarity among the samples in the support dataset and the samples in the query to recognize variety. Mobile-CS architecture was also employed as the backbone network during meta-learning. MobileNetV2 network structure was then enhanced to lighten the original network. A bottleneck structure was also removed to integrate the CBAM attention mechanism. Precise identification of varieties was realized under sample-limited conditions, and then rapid adaptation in new variety identification tasks. An image dataset of wine grape leaves was constructed using 30 varieties, with a total of 5 908 raw images in fields. The experimental results demonstrate that the Deeplabv3+ model achieved an average intersection ratio of 97.52% and a pixel accuracy of 98.98% for precise leaf segmentation. In limited data samples, the two-stage model achieved an average accuracy of 62.27% on the 5-way 1-shot task and 80.06% on the 5-way 5-shot task, indicating superior performance, compared with the rest few-sample learning. Furthermore, the backbone network performed the best with a smaller number of parameters, compared with the classical convolutional neural network. The interference of complex background was also removed on the few-sample learning model. The accuracy of the dataset after the foreground extraction stage on the 5-way 1-shot task was improved by 11.83 percentage points, compared with the original dataset. Ablation experiments were also conducted to verify the improved model. The performance of the improved model was also enhanced to fuse the attention mechanism after the lightweight treatment of the original MobileNetV2 structure. The generalizability was then verified using a publicly available dataset, Leafsnap. There were 1- and 5-shot accuracies of 74.21% and 87.60%, respectively, on the Leafsnap dataset, indicating their superior generalizability. Finally, the T-SNE visualization was used to qualitatively evaluate the improved model. The Mobile-CS-touched data was then better separated in the low-dimensional space. The two-stage variety identification shared high recognition accuracy and strong generalization. A potential solution can also provide for intelligent identification in the field of agriculture. Practical equipment can also be combined with model testing and optimization in future research. This optimal combination can promote the practical application of the technology in agricultural production.

Ningxia Hui Autonomous Region is a key wine grape production area in China. Obtaining timely and accurate information on the distribution of wine grape plots is crucial for adjusting planting structures and estimating yields. This study addresses the challenge of identifying fragmented and irregularly shaped wine grape cultivation areas by utilizing Sentinel-1 and Sentinel-2 remote sensing images on the Google Earth Engine (GEE) cloud platform. Firstly, the study selected three primary wine grape growing regions in Ningxia Hui Autonomous Region: Yongning County and Xixia District as test areas, and Qingtongxia City as the validation area. Sample points were randomly selected within the test area and divided into training and validation sets at a 7:3 ratio. A total of 120 random sample points were selected within the study area to analyze the temporal variation curves of the normalized difference vegetation index (NDVI) reflectance from 2020 Sentinel-2 imagery, as well as the VV and VH polarization backscatter coefficients from Sentinel-1 imagery, to determine the selection dates for multi-temporal images. Based on this foundation, five categories of feature variables were extracted: spectral features (18), vegetation index features (48), texture features (18), polarization features (14), and terrain features (4). Secondly, the Gini index algorithm based on random forests is used to analyze the importance of all feature variables. Features are added in order of importance, from highest to lowest, to achieve the highest classification accuracy and determine the optimal number of features. This optimal number is then used to construct a selected feature set. Additionally, the cumulative feature values of Sentinel-1 and Sentinel-2 images on different dates are calculated. Select the image date with the highest feature score to construct a single temporal feature set. Finally, seven different feature combination schemes were constructed based on single-temporal and multi-temporal feature sets to identify wine grape plots. The optimal feature combination was selected to explore the impact of multi-source and multi-temporal images on the identification of wine grapes, and a spatial distribution map of wine grapes in the study area was created. The study results indicate that combining multisource remote sensing imagery significantly enhances the accuracy of identifying wine grapes compared to using only Sentinel-1 or Sentinel-2 imagery features. Additionally, using multitemporal feature combinations improves classification outcomes by reducing misclassification and omission errors compared to single-temporal feature combinations. Among the seven feature combination schemes, the optimal feature combination demonstrated the best identification performance. In the test area, it achieved an overall accuracy of 95.46%, a Kappa coefficient of 0.94, a user accuracy of 93.33%, and a producer accuracy of 95.06%. The relative error between the extracted area and the area reported in the statistical yearbook is approximately 9.47 percentage points. In the validation area, using sample points from Qingtongxia City, the results indicated an overall accuracy of 91.89%, a Kappa coefficient of 0.89, a user accuracy of 89.00%, and a producer accuracy of 95.93%. The relative error between the extracted area and the area reported in the statistical yearbook is approximately 8.15 percentage points. This study provides valuable insights into classifying and identifying wine grapes through multi-temporal remote sensing images.