Wheat spike detection has important research significance for production estimation and crop field management. With the development of deep learning-based algorithms, researchers tend to solve the detection task by convolutional neural networks (CNNs). However, traditional CNNs equip with the inductive bias of locality and scale-invariance, which makes it hard to extract global and long-range dependency. In this paper, we propose a Transformer-based network named Multi-Window Swin Transformer (MW-Swin Transformer). Technically, MW-Swin Transformer introduces the ability of feature pyramid network to extract multi-scale features and inherits the characteristic of Swin Transformer that performs self-attention mechanism by window strategy. Moreover, bounding box regression is a crucial step in detection. We propose a Wheat Intersection over Union loss by incorporating the Euclidean distance, area overlapping, and aspect ratio, thereby leading to better detection accuracy. We merge the proposed network and regression loss into a popular detection architecture, fully convolutional one-stage object detection, and name the unified model WheatFormer. Finally, we construct a wheat spike detection dataset (WSD-2022) to evaluate the performance of the proposed methods. The experimental results show that the proposed network outperforms those state-of-the-art algorithms with 0.459 mAP (mean average precision) and 0.918 AP50. It has been proved that our Transformer-based method is effective to handle wheat spike detection under complex field conditions.

No abstract

No abstract

No abstract

No abstract

No abstract

Accurate information of soil macronutrient contents and fertilizer macronutrient contents is the precondition of precision fertilization; however, how to detect soil and fertilizer information rapidly, reliably, and inexpensively remains a great challenge. Visible and near-infrared (VIS/NIR) diffuse reflectance spectroscopy proves to be an effective tool for extensive investigation of soil and fertilizer properties. This study first collected many soil and chemical fertilizer samples and performed both spectral scanning and chemical analysis. During the correlation between the collected VIS/NIR spectra and the measured data, different spectral pretreatment, sample selection, and wavelength optimization methods were applied for improving the accuracy and robustness of the prediction models. After appropriate spectral processing and selection of representative samples, both principal component regression and genetic algorithm (GA) can adequately reduce the number of variables and pick out the characteristic variables, which not only enhanced prediction speed but also greatly improved prediction accuracy. In particular, using GA-based models, organic matter content (OMC), total N and pH value in soil and N, P, and K contents in fertilizer can all be accurately predicted.

Laser-induced breakdown spectroscopy (LIBS) is used to determine the total nitrogen (TN) and total phosphorus (TP) in soil. Quantitative determinations are conducted using the line intensity of the analyte element and element concentration. Calibration models are obtained using ten samples for TN and seven samples for TP. The rest samples are used to validate the results. Strong linear correlations are obtained from the determined TN and TP concentrations. LIBS is a powerful tool for analyzing soil samples to determine nutrient elements by selecting calibration and validation samples with similar matrix composition.