Cold damage is one of the disasters that cause significant loss and irreversible damage in crop production. To avoid yield loss, high-throughput phenotyping can be used to select the crop varieties with cold stress resistance. Nowadays, non-destructive spectral image analysis has become an effective way and is widely used in high-throughput phenotyping, which reflects the structural, physiological, biochemical characteristics and traits of plant structure and composition, plant growth and development processes and outcomes. This study used convolutional neural network (CNN) model to extract spectral features in the visible-near-infrared range to estimate cold damage of corn seedlings. The hyperspectral images of cold treated corn seedlings from five varieties were used as research objects in this study. The spectral range of the images was 450-885nm. Gaussian low-pass filter and the Savitzky-Golay smoothing method combined with the first-order derivative was used to do pre-processing for spectral data. For each corn variety, 3600 pixel samples obtained from the selected region of interests in each variety of corn seedlings were used for the CNN modeling. After the CNN modeling, 400 pixel samples extracted from the hyperspectral images were used as the testing set for each variety. Finally, a 10-layer knot CNN model was determined by analyzing the classification accuracy and computational efficiency. CNN detected the cold damage level of different types of corn seedlings as W22 (41.8 %), BxM (35%), B73 (25.6%), PH207 (20%), Mo17 (14%), which had high correlation with the ranking given by chemical method. The coefficient of correlation between cold damage detection results of CNN and results from chemical method is 0.8219. Therefore, it proves that spectral analysis based on CNN modeling can provide reference for detecting cold damage in corn seedlings.

Rapid and accurate measurement of high-resolution soil total nitrogen (TN) information can promote variable rate fertilization, protect the environment, and ensure crop yields. Many scholars focus on exploring the rapid TN detection methods and corresponding soil sensors based on spectral technology. However, soil spectra are easily disturbed by many factors, especially soil moisture and particle size. Real-time elimination of the interferences of these factors is necessary to improve the accuracy and efficiency of measuring TN concentration in farmlands. Although, many methods can be used to eliminate soil moisture and particle size effects on the estimation of soil parameters using continuum spectra. However, the discrete NIR spectral band data can be completely different in the band attribution with continuum spectra, that is, it does not have continuity in the sense of spectra. Thus, relevant elimination methods of soil moisture and particle size effects on continuum spectra do not apply to the discrete NIR spectral band data. To solve this problem, in this study, moisture absorption correction index (MACI) and particle size correction index (PSCI) methods were proposed to eliminate the interferences of soil moisture and particle size, respectively. Soil moisture interference was decreased by normalizing the original spectral band data into standard spectral band data, on the basis of the strong soil moisture absorption band at 1450 nm. For the PSCI method, characteristic bands of soil particle size were identified to be 1361 and 1870 nm firstly. Next, normalized index N_p, which calculated wavelengths of 1631 and 1870 nm, was proposed to eliminate soil particle size interference on discrete NIR spectral band data. Finally, a new coupled elimination method of soil moisture and particle size interferences on predicting TN concentration through discrete NIR spectral band data was proposed and evaluated. The six discrete spectral bands (1070, 1130, 1245, 1375, 1550, and 1680 nm) used in the on-the-go detector of TN concentration were selected to verify the new method. Field tests showed that the new coupled method had good effects on eliminating interferences of soil moisture and soil particle size.

In the practical application of farmland, the soil organic matter prediction model established by the traditional near-infrared (NIR) spectroscopy is affected by factors such as soil texture, which leads to a serious decline in the accuracy of the model. To improve the robustness and prediction accuracy of the model, a prediction model based on NIR spectroscopy and image fusion is proposed. A 1D-CNN organic matter prediction model (based on NIR spectroscopy) was established using eight characteristic wavelengths of extracted soil organic matter (932 nm, 999 nm, 1083 nm, 1191 nm, 1316 nm, 1356 nm, 1583 nm, and 1626 nm) as spectral information. A 2D -CNN organic matter prediction model was established using soil RGB images as information. Based on the idea of model weight fusion, 1D-CNN and 2D-CNN models are fused. When using small convolutional kernels (three-layer convolutional kernel size: 3*3, 1*1, 1*1) and 1D-CNN:2D-CNN = 6:4, the model has the highest prediction accuracy (R ² = 0.872). The optimal fusion model was embedded into the inspection system. The final laboratory and field testing results are as follows: under laboratory conditions, the detection accuracy R ² of the 1D CNN prediction model, 2D-CNN prediction model, and fusion model are 0.838, 0.781, and 0.869, respectively. The root mean square error is 3.005, 3.546, and 2.678, respectively. The above experimental data indicates that the R ² of the fused model is more accurate compared to the model established with a single information. In the field test, the R ² detection accuracy of 1D-CNN prediction model, 2D-CNN prediction model and fusion model is 0.809, 0.731 and 0.835, respectively. The root mean square errors are 3.466, 3.828 and 2.973, respectively. The results show that the fusion model improves the prediction accuracy and model robustness, and the detection system can meet the needs of soil nutrient detection in farmland.

Soil is a heterogeneous medium that exhibits considerable variability in both spatial and temporal dimensions. Proper management of field variability using variable-rate fertilization (VRF) techniques is essential to maximize crop input–output ratios and resource utilization. Implementing VRF technology on a localized scale is recommended to increase crop yield, decrease input costs, and reduce the negative impact on the surrounding environment. This study assessed the agronomic and environmental viability of implementing VRF during the cultivation of summer maize using an on-the-go detector of soil total nitrogen (STN) to detect STN content in the test fields. A spatial delineation approach was then applied to divide the experimental field into multiple management zones. The amount of fertilizer applied in each zone was determined based on the sensor-detected STN. The analysis of the final yield and economic benefits indicates that plots that adopted VRF treatments attained an average summer maize grain yield of 7275 kg ha⁻¹, outperforming plots that employed uniform-rate fertilization (URF) treatments, which yielded 6713 kg ha⁻¹. Through one-way ANOVA, the yield p values of the two fertilization methods were 6.406 × 10⁻¹⁵, 5.202 × 10⁻¹⁵, 2.497 × 10⁻¹⁵, and 3.199 × 10⁻¹⁵, respectively, indicating that the yield differences between the two fertilization methods were noticeable. This led to an average yield increase of 8.37% ha⁻¹ and a gross profit margin of USD 153 ha⁻¹. In plots in which VRF techniques are utilized, the average nitrogen (N) fertilizer application rate is 627 kg ha⁻¹. In contrast, in plots employing URF methods, the N fertilizer application rate is 750 kg ha⁻¹. The use of N fertilizer was reduced by 16.4%. As a result, there is a reduction in production costs of USD 37.5 ha⁻¹, achieving increased yield while decreasing the amount of applied fertilizer. Moreover, in plots where the VRF method was applied, STN was balanced despite the reduced N application. This observation can be deduced from the variance in summer maize grain yield through various fertilization treatments in a comparative experiment. Future research endeavors should prioritize the resolution of particular constraints by incorporating supplementary soil data, such as phosphorus, potassium, organic matter, and other pertinent variables, to advance and optimize fertilization methodologies.

It is of great significance to obtain soil texture information quickly for the realization of farmland management. Soil with good particle condition can well regulate the needs of plants for water, nutrients, air, and temperature during crop growth, thereby promoting high crop yields. The existing methods of measuring soil texture cannot meet the requirements of time and spatial resolution. For this reason, a vehicle-mounted soil texture detector was designed and developed based on machine vision and soil electrical conductivity devices. The detector does not require pretreatment such as air-drying and screening of the soil, and completely uses the original information of the farmland. The whole process can obtain the soil texture information in real time, omitting the complicated chemical process, and saving manpower and material resources. The vehicle-mounted detector is divided into a mechanical part, a control part, and a display part. The mechanical part provides measurement support for the acquisition of soil texture information; the control part collects and processes signals and images; the measurement results can be intuitively observed and recorded on the display, and can be operated through the mobile phone. The vehicle-mounted detector obtains soil conductivity through 4 disc electrodes, while the vehicle-mounted industrial camera captures the soil surface image, and extracts texture parameters through image processing, takes EC and texture parameters as input, and the embedded SVM model of the instrument was used to perform soil texture prediction. In order to verify the measurement accuracy of the detector, farmland verification experiments were carried out on farmland loam in Tongzhou District and Haidian District of Beijing. The R² of the correlation between the measured value of soil EC and the actual value was 0.75, and the accuracy of soil texture prediction was 84.86%. It shows that the developed vehicle-mounted soil texture detector can meet the requirements for rapid acquisition of farmland texture information.