Scab, frogeye spot, and cedar rust are three common types of apple leaf diseases, and the rapid diagnosis and accurate identification of them play an important role in the development of apple production. In this work, an improved model based on VGG16 is proposed to identify apple leaf diseases, in which the global average poling layer is used to replace the fully connected layer to reduce the parameters and a batch normalization layer is added to improve the convergence speed. A transfer learning strategy is used to avoid a long training time. The experimental results show that the overall accuracy of apple leaf classification based on the proposed model can reach 99.01%. Compared with the classical VGG16, the model parameters are reduced by 89%, the recognition accuracy is improved by 6.3%, and the training time is reduced to 0.56% of that of the original model. Therefore, the deep convolutional neural network model proposed in this work provides a better solution for the identification of apple leaf diseases with higher accuracy and a faster convergence speed.

The detection and counting of wheat ears are very important for crop field management, yield estimation, and phenotypic analysis. Previous studies have shown that most methods for detecting wheat ears were based on shallow features such as color and texture extracted by machine learning methods, which have obtained good results. However, due to the lack of robustness of these features, it was difficult for the above-mentioned methods to meet the detection and counting of wheat ears in natural scenes. Other studies have shown that convolutional neural network (CNN) methods could be used to achieve wheat ear detection and counting. However, the adhesion and occlusion of wheat ears limit the accuracy of detection. Therefore, to improve the accuracy of wheat ear detection and counting in the field, an improved YOLOv4 (you only look once v4) with CBAM (convolutional block attention module) including spatial and channel attention model was proposed that could enhance the feature extraction capabilities of the network by adding receptive field modules. In addition, to improve the generalization ability of the model, not only local wheat data (WD), but also two public data sets (WEDD and GWHDD) were used to construct the training set, the validation set, and the test set. The results showed that the model could effectively overcome the noise in the field environment and realize accurate detection and counting of wheat ears with different density distributions. The average accuracy of wheat ear detection was 94%, 96.04%, and 93.11%. Moreover, the wheat ears were counted on 60 wheat images. The results showed that R² = 0.8968 for WD, 0.955 for WEDD, and 0.9884 for GWHDD. In short, the CBAM-YOLOv4 model could meet the actual requirements of wheat ear detection and counting, which provided technical support for other high-throughput parameters of the extraction of crops.

Nitrogen (N) content is an important basis for the precise management of wheat fields. The application of unmanned aerial vehicles (UAVs) in agriculture provides an easier and faster way to monitor nitrogen content. Previous studies have shown that the features acquired from UAVs yield favorable results in monitoring wheat growth. However, since most of them are based on different vegetation indices, it is difficult to meet the requirements of accurate image interpretation. Moreover, resampling also easily ignores the structural features of the image information itself. Therefore, a spectral-spatial feature is proposed combining vegetation indices (VIs) and wavelet features (WFs), especially the acquisition of wavelet features from the UAV image, which was transformed from the spatial domain to the frequency domain with a wavelet transformation. In this way, the complete spatial information of different scales can be obtained to realize good frequency localization, scale transformation, and directional change. The different models based on different features were compared, including partial least squares regression (PLSR), support vector regression (SVR), and particle swarm optimization-SVR (PSO-SVR). The results showed that the accuracy of the model based on the spectral-spatial feature by combining VIs and WFs was higher than that of VIs or WF indices alone. The performance of PSO-SVR was the best (R² = 0.9025, root mean square error (RMSE) = 0.3287) among the three regression algorithms regardless of the use of all the original features or the combination features. Our results implied that our proposed method could improve the estimation accuracy of aboveground nitrogen content of winter wheat from UAVs with consumer digital cameras, which have greater application potential in predicting other growth parameters.

For the issue of low accuracy and poor real-time performance of insulator and defect detection by an unmanned aerial vehicle (UAV) in the process of power inspection, an insulator detection model MobileNet_CenterNet was proposed in this study. First, the lightweight network MobileNet V1 was used to replace the feature extraction network Resnet-50 of the original model, aiming to ensure the detection accuracy of the model while speeding up its detection speed. Second, a spatial and channel attention mechanism convolutional block attention module (CBAM) was introduced in CenterNet, aiming to improve the prediction accuracy of small target insulator position information. Then, three transposed convolution modules were added for upsampling, aiming to better restore the semantic information and position information of the image. Finally, the insulator dataset (ID) constructed by ourselves and the public dataset (CPLID) were used for model training and validation, aiming to improve the generalization ability of the model. The experimental results showed that compared with the CenterNet model, MobileNet_CenterNet improved the detection accuracy by 12.2%, the inference speed by 1.1 f/s for FPS-CPU and 4.9 f/s for FPS-GPU, and the model size was reduced by 37 MB. Compared with other models, our proposed model improved both detection accuracy and inference speed, indicating that the MobileNet_CenterNet model had better real-time performance and robustness.

Nitrogen is an important indicator for monitoring wheat growth. The rapid development and wide application of non-destructive detection provide many approaches for estimating leaf nitrogen content (LNC) in wheat. Previous studies have shown that better results have been obtained in the estimation of LNC in wheat based on spectral features. However, the lack of automatically extracted features leads to poor universality of the estimation model. Therefore, a feature fusion method for estimating LNC in wheat by combining spectral features with deep features (spatial features) was proposed. The deep features were automatically obtained with a convolutional neural network model based on the PyTorch framework. The spectral features were obtained using spectral information including position features (PFs) and vegetation indices (VIs). Different models based on feature combination for evaluating LNC in wheat were constructed: partial least squares regression (PLS), gradient boosting decision tree (GBDT), and support vector regression (SVR). The results indicate that the model based on the fusion feature from near-ground hyperspectral imagery has good estimation effect. In particular, the estimation accuracy of the GBDT model is the best (R² = 0.975 for calibration set, R² = 0.861 for validation set). These findings demonstrate that the approach proposed in this study improved the estimation performance of LNC in wheat, which could provide technical support in wheat growth monitoring.

Tea polyphenols are important ingredients for evaluating tea quality. The rapid development of sensors provides an efficient method for nondestructive detection of tea polyphenols. Previous studies have shown that features obtained from single or multiple sensors yield better results in detecting interior tea quality. However, due to their lack of external features, it is difficult to meet the general evaluation model for the quality of the interior and exterior of tea. In addition, some features do not fully reflect the sensor signals of tea for several categories. Therefore, a feature fusion method based on time and frequency domains from electronic nose (E-nose) and hyperspectral imagery (HSI) is proposed to estimate the polyphenol content of tea for cross-category evaluation. The random forest and the gradient boosting decision tree (GBDT) are used to evaluate the feature importance to obtain the optimized features. Three models based on different features for cross-category tea (black tea, green tea, and yellow tea) were compared, including grid support vector regression (Grid-SVR), random forest (RF), and extreme gradient boosting (XGBoost). The results show that the accuracy of fusion features based on the time and frequency domain from the electronic nose and hyperspectral image system is higher than that of the features from single sensor. Whether based on all original features or optimized features, the performance of XGBoost is the best among the three regression algorithms (R² = 0.998, RMSE = 0.434). Results indicate that the proposed method in this study can improve the estimation accuracy of tea polyphenol content for cross-category evaluation, which provides a technical basis for predicting other components of tea.