Rapid determination of chlorophyll content is significant for evaluating cotton’s nutritional and physiological status. Hyperspectral technology equipped with multivariate analysis methods has been widely used for chlorophyll content detection. However, the model developed on one batch or variety cannot produce the same effect for another due to variations, such as samples and measurement conditions. Considering that it is costly to establish models for each batch or variety, the feasibility of using spectral preprocessing combined with deep transfer learning for model transfer was explored. Seven different spectral preprocessing methods were discussed, and a self-designed convolutional neural network (CNN) was developed to build models and conduct transfer tasks by fine-tuning. The approach combined first-derivative (FD) and standard normal variate transformation (SNV) was chosen as the best pretreatment. For the dataset of the target domain, fine-tuned CNN based on spectra processed by FD + SNV outperformed conventional partial least squares (PLS) and squares-support vector machine regression (SVR). Although the performance of fine-tuned CNN with a smaller dataset was slightly lower, it was still better than conventional models and achieved satisfactory results. Ensemble preprocessing combined with deep transfer learning could be an effective approach to estimate the chlorophyll content between different cotton varieties, offering a new possibility for evaluating the nutritional status of cotton in the field.

Leaf nitrogen concentration (LNC) is a critical indicator of crop nutrient status. In this study, the feasibility of using visible and near-infrared spectroscopy combined with deep learning to estimate LNC in cotton leaves was explored. The samples were collected from cotton’s whole growth cycle, and the spectra were from different measurement environments. The random frog (RF), weighted partial least squares regression (WPLS), and saliency map were used for characteristic wavelength selection. Qualitative models (partial least squares discriminant analysis (PLS-DA), support vector machine for classification (SVC), convolutional neural network classification (CNNC) and quantitative models (partial least squares regression (PLSR), support vector machine for regression (SVR), convolutional neural network regression (CNNR)) were established based on the full spectra and characteristic wavelengths. Satisfactory results were obtained by models based on CNN. The classification accuracy of leaves in three different LNC ranges was up to 83.34%, and the root mean square error of prediction (RMSEP) of quantitative prediction models of cotton leaves was as low as 3.36. In addition, the identification of cotton leaves based on the predicted LNC also achieved good results. These results indicated that the nitrogen content of cotton leaves could be effectively detected by deep learning and visible and near-infrared spectroscopy, which has great potential for real-world application.

Verticillium wilt (VW) is often referred to as the cancer of cotton and it has a detrimental effect on cotton yield and quality. Since the root system is the first to be infested, it is feasible to detect VW by root analysis in the early stages of the disease. In recent years, with the update of computing equipment and the emergence of large-scale high-quality data sets, deep learning has achieved remarkable results in computer vision tasks. However, in some specific areas, such as cotton root MRI image task processing, it will bring some challenges. For example, the data imbalance problem (there is a serious imbalance between the cotton root and the background in the segmentation task) makes it difficult for existing algorithms to segment the target. In this paper, we proposed two new methods to solve these problems. The effectiveness of the algorithms was verified by experimental results. The results showed that the new segmentation model improved the Dice and mIoU by 46% and 44% compared with the original model. And this model could segment MRI images of rapeseed root cross-sections well with good robustness and scalability. The new classification model improved the accuracy by 34.9% over the original model. The recall score and F1 score increased by 59% and 42%, respectively. The results of this paper indicate that MRI and deep learning have the potential for non-destructive early detection of VW diseases in cotton.