The watermelon breed testing system combining machine vision and hyperspectral imaging was developed to enhance efficiency, reduce labor, and improve consistency in quality evaluation. The system included an automatic image acquisition device, featuring a Gige camera and a weighing platform, to capture watermelon phenotypic data. It used hyperspectral imaging on selected regions of interest to model SSC distribution within the watermelon, with measurements of transverse and longitudinal diameters achieved through minimum external rectangle fitting and rind thickness estimated using Canny edge detection. Convolutional smoothing (SGS) algorithm was used to preprocess the spectral data by combining three algorithms, namely multivariate scattering correction (MSC), standard normal transform (SNV), and unit vector normalisation (UVN), respectively, and then the best preprocessed spectral data were filtered by competitive adaptive reweighting algorithm (CARS), successive projection algorithm (SPA), and one-time combined dimensionality reduction algorithm (CARS+SPA) for feature wavelength screening, and finally the screened spectral data were used to build PLSR models and) analyzed for comparison. The results showed that the system had the highest accuracy of 98.68%, 98.82% and 93.81% for the measured values of transverse diameter, longitudinal diameter and rind thickness of watermelon, respectively, with the root mean square error of 2.43 mm, 2.08 mm and 1.63 mm, respectively, as compared with the manual measurements. The best model for predicting watermelon brix was (SGS+UVN)-CARS-PLSR, with prediction correlation coefficients and of 0.920 4 and 0.912 7, root mean square errors RMSEC and RMSEP of 0.376 0°Brix and 0.466 8°Brix, respectively, and a relative analytical error of prediction RPD of 2.49.

Near-infrared spectroscopy has been widely used for the rapid analysis of the physical and chemical properties of straw, offering advantages such as fast detection speed, high efficiency, and non-destructive testing. However, in practical applications, differences in straw particle size can reduce the detection accuracy and stability of the model. For rice straw samples with varying geometric particle sizes, near-infrared spectra were obtained in the 10 000~4 000 cm^-1 wavelength range. Polynomial fitting (PF) was employed to remove background signals from the spectral data of the straw samples. A one-way analysis of variance (ANOVA) was used to perform statistical testing (P<0.05) on the spectral feature signals of the geometrically sized straw samples after polynomial fitting. Principal component analysis (PCA) and Biplot analysis were employed to assess the contribution of spectral feature signals to the spectral differences among straw samples with different geometric particle sizes. Pearson and Spearman correlation analyses were used to screen for key spectral feature signals. Based on the quadratic polynomial fitting mathematical model, it was assumed that absorbance was composed of particle size-related spectral information (PSRSI) and non-particle size-related spectral information (nPSRSI). A particle size regression correction method (PSRCM) was proposed based on the absorbance and particle size of key spectral feature signals. The results indicated that the spectral feature signals at 5 180 cm^-1 of the geometric particle size straw samples exhibited extremely significant differences (P≤0.05, P≤0.01, P≤0.001) and showed a highly significant correlation with particle size (Pearson's r was 0.96). The quadratic polynomial mathematical model performed optimally, with an average coefficient of determination R² and root mean square error (RMSE) of 0.53 and 0.002 1, respectively. PSRCM effectively corrected the spectral differences caused by characteristic signals at 6 824 cm^-1, 5 360 cm^-1, and 5 180 cm^-1, with correction effects significantly superior to those of standard normal variate transformation (SNV) and multiplicative scatter correction (MSC). The research result can provide a theoretical basis for the transfer correction method of straw near-infrared spectroscopy models, which can be applied without the need for standard samples.

Granulation is a common post-harvest disease of citrus fruits, particularly in pomelos. When this disease occurs, it will affect the taste of pomelo fruits and even make them lose their edible value. Pomelos are large, thick-skinned fruits which are difficult to visually detect their internal granulation. The rapid, non-destructive detection of granulation of pomelos was explored by using near-infrared transmission spectroscopy combined with one-dimensional convolutional neural networks. A self-developed visible-near-infrared full-transmission spectroscopy measurement system was used to collect data from red-flashed pomelos. Data were collected in the 562.03~1 110.16 nm range under short integration mode. Without spectral preprocessing, the Kennard-Stones algorithm divided samples into training and testing sets. Subsequently, a one-dimensional convolutional neural network model for pomelo dryness identification was established. The network had a seven-layer structure. Testing showed that the model achieved a classification accuracy of 98.15% for granulation levels. Compared with traditional models such as backpropagation neural network (BPNN), kernel extreme learning machine (K-ELM), and extreme gradient boosting tree (XGBoost), which used raw spectral data and polynomial smoothing processing (SG) and competitive adaptive reweighted sampling (CARS) to extract feature variables. Finally, data augmentation further validated the model's stability. The proposed convolutional neural network model effectively identified internal granulation without complex preprocessing. It can select effective information during the spectral processing, greatly improving the model identification efficiency. The researh provided a reference for the application of deep learning methods to achieve rapid and accurate non-destructive identification of the internal granulation level of pomelo.

Cucumber downy mildew is a fungal disease that severely threatens cucumber production and quality. Quantitative detection of sporangia and conidiophores is crucial for early disease prevention. However, traditional horizontal bounding box detection methods cannot accurately detect these features due to their diverse morphology and orientations. Therefore, an improved YOLO v8s-OBB detection method was proposed by introducing the convolutional block attention module (CBAM) and the lightweight shared convolution detection head (LSCD) module. The aim was to enhance the detection efficiency and accuracy of sporangia and conidiophores of cucumber downy mildew. By incorporating CBAM, the model's ability to identify key features was enhanced, allowing it to focus more on critical regions in microscopic images and improve the detection of small targets. The LSCD integrated multi-scale features through shared convolution operations, enhancing the model's detection performance for targets of different sizes while reducing computational costs, making it suitable for resource-constrained environments. The rotated bounding box technique accurately captured sporangia and conidiophores' inclination and rotation postures. Experimental results showed that, compared with the original YOLO v8s-OBB model, the improved YOLO v8s-OBB model not only reduced the model size but also achieved superior detection performance for sporangia and conidiophores of cucumber downy mildew, with precision, recall, and mAP@0.5 reaching 96.0%, 90.1%, and 96.5%, respectively. The improved YOLO v8s-OBB model outperformed advanced rotated object detection models such as S2ANet, H2RBox, and R2CNN in detection accuracy. The research result can validate the effectiveness of the improved model in practical applications and provide technical support for the early diagnosis of cucumber downy mildew.

Wheat, as one of the world's essential food crops, requires timely diagnosis and control of diseases and pests to significantly reduce yield losses, which is crucial for global food security. However, deep learning methods typically rely on large amounts of training data and high-performance computing resources, which pose limitations in scenarios with small sample sizes and limited resources. To address these issues, a knowledge-enhanced lightweight multi-modal wheat diseases and pests identification model named multi-modal blend convolutional neural network (Blend-CNN) was proposed. The model was structured around a dual-branch convolutional neural network framework. It utilized an EfficientNet backbone network to extract image features of diseases and pests and incorporated a multi-branch TextCNN backbone network to extract textual features of diseases and pests descriptions, thereby obtaining more feature information and improving identification accuracy. Additionally, an innovative convolutional network-based multi-modal fusion method was introduced, allowing the model to optimally integrate information from both modalities globally. Furthermore, to mitigate the accuracy loss common in traditional multi-modal methods, the gradient blend loss function was developed. Finally, to verify the model's effectiveness, comparative experiments were conducted on a constructed dataset containing 880 samples. The results demonstrated that the proposed model achieved the highest identification accuracy of 96.95% on this dataset, compared with other models, it had fewer parameters, lower complexity, and it was more lightweight which was applicable for edge devices, offering theoretical support for wheat diseases and pests identification in scenarios with limited sample sizes and resources.

Aiming to address the issues of small target size, varying pest postures, and low detection efficiency, and suboptimal performance of trapping devices during live rice pest identification, an intelligent recognition method was proposed based on an improved YOLO v8n model integrated with a one-to-three live targeted trapping device. By optimizing the structural design of the traditional targeted recognition equipment, a one-to-three live targeted trapping device was developed to enhance operational efficiency. For the construction of the dataset required for intelligent recognition, a frame-by-frame image clarity comparison and sorting method was proposed to improve data quality. Additionally, an adaptive module was introduced into the improved CBAM attention mechanism to enhance the model's focus on target regions, and a high-resolution P2 layer was added to the AFPN structure to improve small target recognition capability. Comparative experiments demonstrated that the improved YOLO v8n model achieved a mean average precision (mAP@0.5) of 95.65% on the rice leaf folder dataset. Ablation experiments further verified that the enhanced CBAM module and the introduction of the P2 layer in the AFPN structure significantly improved model performance, resulting in optimal overall recognition capability. Field trials confirmed the practical effectiveness of the improved YOLO v8n model, achieving a recognition rate of over 90% for rice leaf folder pests across 60 mu (approximately 4 hectares) of rice fields when combined with the targeted identification device, demonstrating good robustness and stability. In summary, the intelligent recognition method for live rice pests based on YOLO v8n the improved model proposed effectively enhanced recognition accuracy and efficiency, providing a scientific basis for precise pest identification and control in rice fields, with strong practical application value and promotion potential.

Gossypol, a toxic compound unique to the Gossypium species, holds significant research value in agriculture and pharmaceuticals. Pigment glands, the primary carriers of gossypol, exhibit phenotypic traits that strongly correlate with gossypol content, making them effective indicators for its evaluation. However, the detection of pigment glands primarily relies on manual analysis, which is both time-consuming and inefficient. To address these limitations, Gland-MSConNet, a novel model designed for the accurate segmentation of cotton leaf pigment glands in complex backgrounds was proposed. The model integrated the convolutional bidirectional Mamba (ConBimamba) and multi-scale attention aggregation (MSAA) modules, significantly enhancing its ability to capture multi-scale features and improving robustness in challenging scenarios. Additionally, the incorporation of sub-pixel convolution facilitated efficient upsampling, preserving fine-grained details. Experimental results demonstrated that Gland-MSConNet achieved a segmentation mIoU of 89.83% and a recall rate of 95.03%, representing improvements of 4.66 percentage points and 5.78 percentage points, respectively, over the U-Net model. These findings highlighted the model's high segmentation accuracy and robustness. The Gland-MSConNet model provided a strong technical foundation for the segmentation and phenotypic quantification of cotton leaf pigment glands, offering a reliable basis for the automated evaluation of gossypol content.

Cucumber is one of the most important vegetables and economic crops in the world. The occurrence of fungal diseases in cucumbers seriously threatens the safety of cucumber production, with powdery mildew being one of the most common fungal diseases. With the rapid development of computer technology, more and more deep learning algorithms are being applied to identify powdery mildew fungus. However, existing algorithms suffer from low accuracy in recognizing small and occluded targets, as well as insufficient localization precision. To address this issue, the parallelized patch-aware attention (PPA) module was firstly introduced into the backbone network of YOLO v8s. By employing a parallel multi-branch structure and attention mechanism, it effectively captured multi-scale features of small targets, preserved critical information during multiple downsampling processes, and enhanced the performance of small target detection. Additionally, the global-to-local spatial aggregation (GLSA) module was introduced into the neck, which combined global contextual information with local detail features, significantly improving the model's feature representation capability. This module enhanced the detection performance for small targets and complex scenes by better capturing multi-scale features. Experimental results showed that PG-YOLO v8s significantly improved powdery mildew fungus detection performance compared with YOLO v8s. The network achieved high precision in detecting powdery mildew fungus, with notable improvements in the detection accuracy of small and occluded targets. The research result can provide a high-throughput method for detecting powdery mildew fungus, enabling precise early detection and guiding early intelligent decision-making in cucumber production. This approach can help to improve disease control efficiency, ensure cucumber yield and quality, and it was of great significance for the sustainable development of agricultural production.

Addressing the challenges of apricot mushroom grading, which relies heavily on manual labor, is inefficient, and prone to subjectivity. An automated grading and detection method for apricot mushrooms named FCML-YOLO v8 was proposed. Building upon the improved YOLO v8n-seg, the network sequentially integrated the feature-focused diffusion pyramid network (FDPN), characteristic attention fusion module (CAFM), and multi-level channel attention (MLCA) mechanism. By combining the FCML-YOLO v8 model with subordinate judgment, mask merging, and center screening methods, accurate assessments of apricot mushroom size and disease spot conditions was achieved. Additionally, through image refinement and curve-fitting techniques, it effectively quantified the curvature of apricot mushrooms. Furthermore, by leveraging CIE XYZ color space values, the color characteristics of apricot mushrooms were precisely analyzed. By integrating these four indicators, the automated grading and detection of apricot mushrooms were successfully implemented. Experimental results demonstrated that the FCML-YOLO v8 model achieved a precision of 94.41%, a recall rate of 88.48%, and an mAP@0.5 of 92.04% in bounding box prediction. In mask detection, it achieved a precision of 90.52%, a recall rate of 84.14%, and an mAP@0.5 of 86.79%. Compared with the original YOLO v8n-seg model, the mAP@0.5 of FCML-YOLO v8 was improved by 6.39 and 4.79 percentage points, respectively, in bounding box prediction and mask detection. In the constructed apricot mushroom grading and detection experimental setup, the FCML-YOLO v8 model achieved an mAP@0.5 of 92.98%, fully meeting the requirements of industrial-grade applications.

The integration of multi-source heterogeneous agricultural information not only help understanding and unveiling the pathways of diseases infection and prevention but also offers vital data support for research on diseases prescription recommendations. Aiming at the problems of fusion, alignment and heterogeneity in the analysis of multi-source heterogeneous agricultural data such as electronic medical records, the knowledge graph of crop diseases prescription was constructed, and the diseases association on this basis was visually analyzed. Initially, starting from the principle of the disease triangle, the critical roles of pathogens, hosts, and environment in the process of disease infection and control were analyzed, constructing the ontology layer of the crop diseases prescription knowledge graph with 18 ontology concepts, 17 relationships, and 6 attribute edges based on data characteristics. Subsequently, the entity layer of the knowledge graph was built by integrating rule-matching and deep-learning knowledge extraction methods, encompassing 1 121 entity instances and 8 292 relationship instances. Lastly, identification of Top20 key nodes based on degree and betweenness centrality, along with Top5 disease-prevention product association predictions using the Adamic-Adar index, were conducted to visually analyze the associations between key entities and attributes, including diseases, symptoms, and prevention products. Six rules were established to enhance the recommendation and information retrieval functions of disease prevention products from three perspectives: prevention scheme selection, green subsidy filtering, and related information inquiry. The research result can provide reference for electronic medical record data mining and association analysis of crop diseases.