Citrus has become a pivotal industry for the rapid development of agriculture and increasing farmers’ incomes in the main production areas of southern China. Knowing how to diagnose and control citrus huanglongbing has always been a challenge for fruit farmers. To promptly recognize the diagnosis of citrus huanglongbing, a new classification model of citrus huanglongbing was established based on MobileNetV2 with a convolutional block attention module (CBAM-MobileNetV2) and transfer learning. First, the convolution features were extracted using convolution modules to capture high-level object-based information. Second, an attention module was utilized to capture interesting semantic information. Third, the convolution module and attention module were combined to fuse these two types of information. Last, a new fully connected layer and a softmax layer were established. The collected 751 citrus huanglongbing images, with sizes of 3648 × 2736, were divided into early, middle, and late leaf images with different disease degrees, and were enhanced to 6008 leaf images with sizes of 512 × 512, including 2360 early citrus huanglongbing images, 2024 middle citrus huanglongbing images, and 1624 late citrus huanglongbing images. In total, 80% and 20% of the collected citrus huanglongbing images were assigned to the training set and the test set, respectively. The effects of different transfer learning methods, different model training effects, and initial learning rates on model performance were analyzed. The results show that with the same model and initial learning rate, the transfer learning method of parameter fine tuning was obviously better than the transfer learning method of parameter freezing, and that the recognition accuracy of the test set improved by 1.02~13.6%. The recognition accuracy of the citrus huanglongbing image recognition model based on CBAM-MobileNetV2 and transfer learning was 98.75% at an initial learning rate of 0.001, and the loss value was 0.0748. The accuracy rates of the MobileNetV2, Xception, and InceptionV3 network models were 98.14%, 96.96%, and 97.55%, respectively, and the effect was not as significant as that of CBAM-MobileNetV2. Therefore, based on CBAM-MobileNetV2 and transfer learning, an image recognition model of citrus huanglongbing images with high recognition accuracy could be constructed.

The accurate identification of citrus fruits is important for fruit yield estimation in complex citrus orchards. In this study, the YOLOv7-tiny-BVP network is constructed based on the YOLOv7-tiny network, with citrus fruits as the research object. This network introduces a BiFormer bilevel routing attention mechanism, which replaces regular convolution with GSConv, adds the VoVGSCSP module to the neck network, and replaces the simplified efficient layer aggregation network (ELAN) with partial convolution (PConv) in the backbone network. The improved model significantly reduces the number of model parameters and the model inference time, while maintaining the network’s high recognition rate for citrus fruits. The results showed that the fruit recognition accuracy of the modified model was 97.9% on the test dataset. Compared with the YOLOv7-tiny, the number of parameters and the size of the improved network were reduced by 38.47% and 4.6 MB, respectively. Moreover, the recognition accuracy, frames per second (FPS), and F1 score improved by 0.9, 2.02, and 1%, respectively. The network model proposed in this paper has an accuracy of 97.9% even after the parameters are reduced by 38.47%, and the model size is only 7.7 MB, which provides a new idea for the development of a lightweight target detection model.

Citrus tangerines are famous fruits worldwide, and monitoring the water content of citrus leaves is highly important for citrus production. However, there are still challenges in quantitatively estimating the water content of citrus leaves using hyperspectral technology, and the random noise generated during spectral acquisition and the overlapping peaks in the sensitive band of the citrus leaf water content will affect estimation accuracy. To solve these problems and further explore the roles of the continuous wavelet transform (CWT) and fractional-order derivative (FOD) in the estimation of citrus leaf water content, this study intends to use of CWT and FOD to decompose the original spectrum, and then compare the correlation between the original spectrum and leaf water content to explore whether the decomposition treatment has improved the correlation between spectrum and leaf moisture content. Then, the successive projections algorithm (SPA) was used to select feature bands and combine spectral vegetation indices. Partial least squares regression (PLSR) was used to construct water-content inversion models for citrus leaves, and the inversion accuracies of two commonly used spectral preprocessing methods were compared. The results indicate that (1) the CWT can improve the sensitivity of the spectrum to the citrus leaf water content to a certain extent, and the inversion accuracy of the CWT is approximately 5% greater than that of the FOD. (2) On the basis of the CWT and FOD methods, the inversion accuracy of the citrus leaf water content based on SPA screening increased by 9.61% and 9.29%, respectively, compared with the original spectrum. (3) Under CWT decomposition, Scale4 of the Gaus1 wavelet was screened by the SPA, and the inversion model of citrus leaf water content was constructed by combining the spectral vegetation index NDVI with the best results. The R-squared (${\mathrm{R}}^2$) and root mean square error (RMSE) values were 0.7491 and 0.0284, respectively, which were both 0.0138 greater than those of the best inversion model for the FOD ${\mathrm{R}}^2$. In conclusion, the CWT-SPA combined with the spectral vegetation index can improve the sensitivity of the spectrum to the citrus leaf water content, eliminate a large amount of redundant data, and enhance the prediction ability and stability of the citrus leaf water content.

To address the current issues of complex structures and low accuracies in citrus pest identification models, a lightweight pest identification model was proposed. First, a parameterized linear rectification function was introduced to avoid neuronal death. Second, the model’s attention to pest characteristics was improved by incorporating an improved mixed attention mechanism. Subsequently, the network structure of the original model was adjusted to reduce architectural complexity. Finally, by employing transfer learning, an SCHNet model was developed. The experimental results indicated that the proposed model achieved an accuracy rate of 94.48% with a compact size of 3.84 MB. Compared to the original ShuffleNet V2 network, the SCHNet model showed a 3.12% accuracy improvement while reducing the model size by 22.7%. The SCHNet model exhibited an excellent classification performance for citrus pest identification, enabling the accurate identification of citrus pests.

Rapid and nondestructive prediction of chlorophyll content and response to the growth of various crops using remote sensing technology is a prominent topic in agricultural remote sensing research. Bordeaux mixture has been extensively employed for managing citrus diseases, such as black star and ulcer disease. However, the presence of pesticide residues in Bordeaux mixture can significantly modify the spectral response of the citrus canopy, thereby exerting a substantial influence on the accurate prediction of agronomic indices in fruit trees. In this study, we used unmanned aerial vehicle (UAV) multispectral imaging technology to obtain remote sensing imagery of Bordeaux-covered citrus canopies during the months of July, September, and November. We integrated spectral and texture information to construct a high-dimensional feature dataset and performed data downscaling and feature optimization. Furthermore, we established four machine learning models, namely, partial least squares regression (PLS), ridge regression (RR), ridge, random forest (RF), and support vector regression (SVR). Our objectives were to identify the most effective prediction model for estimating the SPAD (soil plant analysis development) value of Bordeaux-covered citrus canopies, assess the variation in prediction accuracy between fused features and individual features, and investigate the impact of Bordeaux solution on the spectral reflectance of the citrus canopy. The results showed that (1) the impact of Bordeaux mixture on citrus canopy reflectance bands ranked from the highest to the lowest as follows: near-infrared band at 840 nm, red-edge band at 730 nm, blue band at 450 nm, green band at 560 nm, and red band at 650 nm. (2) Fused feature models had better prediction ability than single-feature modeling, with an average R² value of 0.641 for the four model test sets, improving by 0.117 and 0.039, respectively, compared with single-TF (texture feature) and -VI (vegetation index) modeling, and the test-set root-mean-square error (RMSE) was 2.594 on average, which was 0.533 and 0.264 lower than single-TF and -VI modeling, respectively. (3) Multiperiod data fusion effectively enhanced the correlation between features and SPAD values and consequently improved model prediction accuracy. Compared with accuracy based on individual months, R improved by 0.013 and 0.011, while RMSE decreased by 0.112 and 0.305. (4) The SVR model demonstrated the best performance in predicting citrus canopy SPAD under Bordeaux solution coverage, with R² values of 0.629 and 0.658, and RMSE values of 2.722 and 2.752 for the training and test sets, respectively.

Image segmentation is an important task in image processing and analysis but due to the same ground object having different spectra and different ground objects having similar spectra, segmentation, particularly on high-resolution remote sensing images, can be significantly challenging. Since the spectral distribution of high-resolution remote sensing images can have complex characteristics (e.g., asymmetric or heavy-tailed), an innovative image segmentation algorithm is proposed based on the hierarchical Student’s-t mixture model (HSMM) and spatial constraints with adaptive smoothing. Considering the complex distribution of spectral intensities, the proposed algorithm constructs the HSMM to accurately build the statistical model of the image, making more reasonable use of the spectral information and improving segmentation accuracy. The component weight is defined by the attribute probability of neighborhood pixels to overcome the influence of image noise and make a simple and easy-to-implement structure. To avoid the effects of artificially setting the smoothing coefficient, the gradient optimization method is used to solve the model parameters, and the smoothing coefficient is optimized through iterations. The experimental results suggest that the proposed HSMM can accurately model asymmetric, heavy-tailed, and bimodal distributions. Compared with traditional segmentation algorithms, the proposed algorithm can effectively overcome noise and generate more accurate segmentation results for high-resolution remote sensing images.