Detection of the Fusarium head blight (FHB) is crucial for wheat yield protection, with precise and rapid FHB detection increasing wheat yield and protecting the agricultural ecological environment. FHB detection tasks in agricultural production are currently handled by cloud servers and utilize unmanned aerial vehicles (UAVs). Hence, this paper proposed a lightweight model for wheat ear FHB detection based on UAV-enabled edge computing, aiming to achieve the purpose of intelligent prevention and control of agricultural disease. Our model utilized the You Only Look Once version 4 (YOLOv4) and MobileNet deep learning architectures and was applicable in edge devices, balancing accuracy, and FHB detection in real-time. Specifically, the backbone network Cross Stage Partial Darknet53 (CSPDarknet53) of YOLOv4 was replaced by a lightweight network, significantly decreasing the network parameters and the computing complexity. Additionally, we employed the Complete Intersection over Union (CIoU) and Non-Maximum Suppression (NMS) to regress the loss function to guarantee the detection accuracy of FHB. Furthermore, the loss function incorporated the focal loss to reduce the error caused by the unbalanced positive and negative sample distribution. Finally, mixed-up and transfer learning schemes enhanced the model’s generalization ability. The experimental results demonstrated that the proposed model performed admirably well in detecting FHB of the wheat ear, with an accuracy of 93.69%, and it was somewhat better than the MobileNetv2-YOLOv4 model (F1 by 4%, AP by 3.5%, Recall by 4.1%, and Precision by 1.6%). Meanwhile, the suggested model was scaled down to a fifth of the size of the state-of-the-art object detection models. Overall, the proposed model could be deployed on UAVs so that wheat ear FHB detection results could be sent back to the end-users to intelligently decide in time, promoting the intelligent control of agricultural disease.

Aboveground biomass (AGB) is an important basis for wheat yield formation. It is useful to timely collect the AGB data to monitor wheat growth and to build high-yielding wheat groups. However, as traditional AGB data acquisition relies on destructive sampling, it is difficult to adapt to the modernization of agriculture, and the estimation accuracy of spectral data alone is low and cannot solve the problem of index saturation at later stages. In this study, an unmanned aerial vehicle (UAV) with an RGB camera and the real-time kinematic (RTK) was used to obtain imagery data and elevation data at the same time during the critical fertility period of wheat. The cumulative percentile and the mean value methods were then used to extract the wheat plant height (PH), and the color indices (CIS) and PH were combined to invert the AGB of wheat using parametric and non-parametric models. The results showed that the accuracy of the model improved with the addition of elevation data, and the model with the highest accuracy of multi-fertility period estimation was PLSR (PH + CIS), with R², RMSE and NRMSE of 0.81, 1248.48 kg/ha and 21.77%, respectively. Compared to the parametric models, the non-parametric models incorporating PH and CIS greatly improved the prediction of AGB during critical fertility periods in wheat. The inclusion of elevation data therefore greatly improves the accuracy of AGB prediction in wheat compared to traditional spectral prediction models. The fusion of UAV-based elevation data and image information provides a new technical tool for multi-season wheat AGB monitoring.

The number of maize seedlings is a key determinant of maize yield. Thus, timely, accurate estimation of seedlings helps optimize and adjust field management measures. Differentiating “multiple seedlings in a single hole” of maize accurately using deep learning and object detection methods presents challenges that hinder effectiveness. Multivariate regression techniques prove more suitable in such cases, yet the presence of weeds considerably affects regression estimation accuracy. Therefore, this paper proposes a maize and weed identification method that combines shape features with threshold skeleton clustering to mitigate the impact of weeds on maize counting. The threshold skeleton method (TS) ensured that the accuracy and precision values of eliminating weeds exceeded 97% and that the missed inspection rate and misunderstanding rate did not exceed 6%, which is a significant improvement compared with traditional methods. Multi-image characteristics of the maize coverage, maize seedling edge pixel percentage, maize skeleton characteristic pixel percentage, and connecting domain features gradually returned to maize seedlings. After applying the TS method to remove weeds, the estimated R² is 0.83, RMSE is 1.43, MAE is 1.05, and the overall counting accuracy is 99.2%. The weed segmentation method proposed in this paper can adapt to various seedling conditions. Under different emergence conditions, the estimated R² of seedling count reaches a maximum of 0.88, with an RMSE below 1.29. The proposed approach in this study shows improved weed recognition accuracy on drone images compared to conventional image processing methods. It exhibits strong adaptability and stability, enhancing maize counting accuracy even in the presence of weeds.

The reduction of nitrogen (N) fertilizer during the rice growing season is an important practice in rice production and is ecologically beneficial. Will different N reduction stages affect rice yields and NUE? The timing of the reduction in N-efficient varieties (NEVs) is yet to be identified, especially under moderate N rate applications. We investigated the effectiveness of various N reduction stages (NRSs) on grain yield and N-use efficiency (NUE) in NEVs in the lower reaches of the Yangtze River in China. Two NEVs were grown in the field, and five N reduction treatments, including basal N reduction (BR) at pre-transplanting (PT), tillering N reduction (TR) at early tillering (ET), promoting-spikelet N reduction (PR) at panicle initiation (PI), keeping-spikelet N reduction (KR) at spikelet differentiation (SD), and N split reduction (SR) at all four stages, were adopted, with no N reduction (CK) and no N application (N0) as controls. The results showed that grain yield and NUE varied substantially with the NRSs. Yield decreases were observed in descending order of magnitude in BR, PR, SR, TR, and KR when compared to CK. For both NEVs, BR and PR were the most effective treatments in decreasing yield and NUE at the same N reduction rate. BR and PR markedly decreased the panicles per unit area or spikelets per panicle, root biomass, root length, root length density, and root oxidation activity and exhibited simultaneously decreased leaf area index, grain leaf ratio, shoot biomass, and crop growth rate from joining to the heading and from heading to maturity. According to the results, PT and PI were considered to be N reduction sensitive stages, and ET and SD were considered to be N reduction insensitive stages. According to the results, an N reduction strategy was suggested as follows: N reduction at SD and ET, with increased N proportions at PT and PI for NEVs when adopting moderate N application rates.

The number of cultivated wheat seedlings per unit area allows calculation of plant density. Wheat seedling density provides emergence data and this is useful for improving crop management. The number of wheat seedlings is typically determined by visual counts but this is time-consuming and laborious. We obtained field digital images of 1st to 3rd leaf stage wheat seedlings. The seedlings were extracted using an image analysis technique that calculated the coverage degree of the seedlings and the number of angular points of overlapping leaves. The wheat seedling quantity estimation model was constructed using multivariate regression analysis. The model parameters included coverage degree, number of angular points, variety coefficient, and leaf age. Introduction of the number of angular points increased the accuracy of the single coverage degree model. The R2 value was consistently > 0.95 when the model was applied to different varieties, indicating that the model was adaptable for different varieties. As the leaf stage or density increased, the accuracy of the model declined, but the minimum R2 remained > 0.87, indicating good adaptability of the model to seedlings with different leaf ages and densities. This method is an effective means for counting wheat seedlings in the 1st to the 3rd leaf stages.

Nitrogen (N) management is of great importance in rice production, but most previous studies have focused on high N rates and there is a lack of research on management plans under a moderate N rate. This study aimed to explore the agronomic and physiological traits of N-efficient rice varieties (NEVs) and to optimize the management strategy at an N rate below the inflection point of the parabolic curve between N rate and grain yield. Two NEVs and two N-inefficient rice varieties (NIVs) were planted, and three treatments were designed according to the panicle N application method. A larger amount of N applied at panicle initiation (PI) led to higher rice yield and N-use efficiency (NUE). This was mainly due to increases in the total number of spikelets per unit area, root oxidation activity, leaf area duration, and leaf photosynthesis rate as well as to the increased carbon (C) and N utilization rates. Compared with NIVs, NEVs exhibited improved root and shoot functions and higher C and N transport characteristics at the moderate N rate. We suggest that increasing the application of N at PI and that planting of NEVs are important ways to increase rice yield and NUE when adopting moderate N rates.