Crop leaf diseases can reflect the current health status of the crop, and the rapid and automatic detection of field diseases has become one of the difficulties in the process of industrialization of agriculture. In the widespread application of various machine learning techniques, recognition time consumption and accuracy remain the main challenges in moving agriculture toward industrialization. This article proposes a novel network architecture called YOLO V5-CAcT to identify crop diseases. The fast and efficient lightweight YOLO V5 is chosen as the base network. Repeated Augmentation, FocalLoss, and SmoothBCE strategies improve the model robustness and combat the positive and negative sample ratio imbalance problem. Early Stopping is used to improve the convergence of the model. We use two technical routes of model pruning, knowledge distillation and memory activation parameter compression ActNN for model training and identification under different hardware conditions. Finally, we use simplified operators with INT8 quantization for further optimization and deployment in the deep learning inference platform NCNN to form an industrial-grade solution. In addition, some samples from the Plant Village and AI Challenger datasets were applied to build our dataset. The average recognition accuracy of 94.24% was achieved in images of 59 crop disease categories for 10 crop species, with an average inference time of 1.563 ms per sample and model size of only 2 MB, reducing the model size by 88% and the inference time by 72% compared with the original model, with significant performance advantages. Therefore, this study can provide a solid theoretical basis for solving the common problems in current agricultural disease image detection. At the same time, the advantages in terms of accuracy and computational cost can meet the needs of agricultural industrialization.

To solve the problems of weak generalization of potato early and late blight recognition models in real complex scenarios, susceptibility to interference from crop varieties, colour characteristics, leaf spot shapes, disease cycles and environmental factors, and strong dependence on storage and computational resources, an improved YOLO v5 model (DA-ActNN-YOLOV5) is proposed to study potato diseases of different cycles in multiple regional scenarios. Thirteen data augmentation techniques were used to expand the data to improve model generalization and prevent overfitting; potato leaves were extracted by YOLO v5 image segmentation and labelled with LabelMe for building data samples; the component modules of the YOLO v5 network were replaced using model compression technology (ActNN) for potato disease detection when the device is low on memory. Based on this, the features extracted from all network layers are visualized, and the extraction of features from each network layer can be distinguished, from which an understanding of the feature learning behavior of the deep model can be obtained. The results show that in the scenario of multiple complex factors interacting, the identification accuracy of early and late potato blight in this study reached 99.81%. The introduced data augmentation technique improved the average accuracy by 9.22%. Compared with the uncompressed YOLO v5 model, the integrated ActNN runs more efficiently, the accuracy loss due to compressed parameters is less than 0.65%, and the time consumption does not exceed 30 min, which saves a lot of computational cost and time. In summary, this research method can accurately identify potato early and late blight in various scenarios.

In China, fruit tree diseases are a significant threat to the development of the fruit tree industry, and knowledge about fruit tree diseases is the most needed professional knowledge for fruit farmers and other practitioners in the fruit tree industry. Research papers are the primary sources of professional knowledge that represent the cutting-edge progress in fruit disease research. Traditional knowledge engineering methods for knowledge acquisition require extensive and cumbersome preparatory work, and they demand a high level of professional background and information technology skills from the handlers. This paper, from the perspective of fruit tree industry knowledge dissemination, aims at users such as fruit farmers, fruit tree experts, fruit tree knowledge communicators, and information gatherers. It proposes a fast, cost-effective, and low-technical-barrier method for extracting fruit tree disease knowledge from research paper abstracts—K-Extract, based on large language models (LLMs) and prompt engineering. Under zero-shot conditions, K-Extract utilizes conversational LLMs to automate the extraction of fruit tree disease knowledge. The K-Extract method has constructed a comprehensive classification system for fruit tree diseases and, through a series of optimized prompt questions, effectively overcomes the deficiencies of LLM models in providing factual accuracy. This paper tests multiple LLM models available in the Chinese market, and the results show that K-Extract can seamlessly integrate with any conversational LLM model, with the DeepSeek model and the Kimi model performing particularly well. The experimental results indicate that LLM models have a high accuracy rate in handling judgment tasks and simple knowledge Q&A tasks. The K-Extract method is simple, efficient, and accurate, and can serve as a convenient tool for knowledge extraction in the agricultural field.

Accurate diagnosis of plant diseases is crucial for crop health. This study introduces the EDA–ViT model, a Vision Transformer (ViT)-based approach that integrates adaptive entropy-based data augmentation for diagnosing custard apple (Annona squamosa) diseases. Traditional models like convolutional neural network and ViT face challenges with local feature extraction and large dataset requirements. EDA–ViT overcomes these by using a multi-scale weighted feature aggregation and a feature interaction module, enhancing both local and global feature extraction. The adaptive data augmentation method refines the training process, boosting accuracy and robustness. With a dataset of 8226 images, EDA–ViT achieved a classification accuracy of 96.58%, an F1 score of 96.10%, and a Matthews Correlation Coefficient (MCC) of 92.24%, outperforming other models. The inclusion of the Deformable Multi-head Self-Attention (DMSA) mechanism further enhanced feature capture. Ablation studies revealed that the adaptive augmentation contributed to a 0.56% accuracy improvement and a 0.34% increase in MCC. In summary, EDA–ViT presents an innovative solution for custard apple disease diagnosis, with potential applications in broader agricultural disease detection, ultimately aiding precision agriculture and crop health management.