Pests significantly threaten global agricultural production, which causes severe yield losses through feeding and virus transmission. To mitigate yield losses caused by pests, timely and precise pest management practices are critical. Although previous efforts have advanced automated solutions for real-time environmental monitoring in agriculture, implementing precise pest management decision-making and suggestion generation remains a challenge due to complex reasoning processes in practice. In response, an enhanced pest management system, PEZEGO, is proposed to provide precise management suggestions through multimodal environmental data, a fine-tuned open vocabulary detector (OVD), and large language models (LLMs). Specifically, a mobile application and low-cost Internet of Things (IoT) devices are developed to capture images and environmental information. A hybrid convolutional low-rank adaptation method (HCLoRA) is proposed to fine-tune pretrained OVDs, enabling zero-shot pest detection for converting images to pest species and quantity information. In addition, a structured data-based retrieval augmented generation (SRAG) workflow for LLMs is proposed to provide precise pest management suggestions through automatically extracted agriculture management knowledge and Chain-of-Thought. The effectiveness of PEZEGO is validated in a case study of pest management in the U.K., including pest detection in field scenarios and management suggestion generation. Compared to advanced model fine-tuning methods, HCLoRA for YOLOWorld achieves the highest detection performance with$0.1759~AP^{h}$on pest detection. Additionally, the proposed SRAG workflow demonstrates the ability to support pest management with a 68.7% average F1 score for knowledge extraction and 77.33% accuracy for suggestion generation. Eventually, a mobile application demonstrates the practical effectiveness of the proposed system.

With the global ageing population, the diagnosis of Alzheimer's disease (AD) has become an urgent public health priority. Gene expression techniques offer the advantages of being less invasive and cost-effective, but their high dimensionality and small sample sizes make them prone to the curse of dimensionality in AD diagnosis. This study proposes a novel interpretable deep learning approach to address these challenges. We introduce a shallow sparse autoencoder for dimensionality reduction and combine it with XGBoost for classification, achieving an Area Under the Receiver Operating Characteristic curve (AUROC) of up to 95.13% . Additionally, we develop a fast, low-cost feature selection algorithm that dynamically adjusts feature elimination to enhance model efficiency. Comprehensive cross-dataset evaluation demonstrates the model's strong generalisation performance on the public datasets: Alzheimer's Disease Neuroimaging Initiative (ADNI), AddNeuroMed1 (ANM1), and ANM2. Our method also provides biological interpretability through enrichment analysis, offering insights into the mechanisms underlying AD and potential therapeutic targets. This makes our approach a promising tool for early, accurate diagnosis and clinical application.

In the rapidly evolving field of artificial intelligence, large language models (LLMs) have attracted much attention from researchers in various fields due to their unexpected text generation and comprehension capabilities. However, the applications of LLMs for sustainable pest management are under-explored due to the heavy reliance on specialized expert knowledge. In addition, evaluating the quality of LLMs' content is another technological challenge for applying LLMs in sustainable pest management. Therefore, we propose an instruction-based prompting method that integrates pest expert knowledge into the prompt, equipping LLMs with the necessary context to generate more accurate and relevant pest management advice. Furthermore, we propose an LLM-based evaluation framework to score the generated content on Coherence, Logical Consistency, Fluency, Relevance, Comprehension, and Exhaustion. Additionally, we integrate an Expert System based on crop threshold data as a baseline to obtain scores for Accuracy on whether pests found in crop fields should take management action. Each model's score is weighted by percentage to get a final score. The results show that GPT-3.5 and GPT-4 outperform the FLAN models in most evaluation dimensions. Furthermore, while using instruction-based prompting containing domain-specific knowledge outperforms other prompting methods with an accuracy of 72%, ongoing refinements and assessments of end-user satisfaction are essential to enhance the LLMs' effectiveness and practical helpfulness in providing pest management advice.

Multimodal learning is widely used in automated early diagnosis of Alzheimer's disease. However, the current studies are based on an assumption that different modalities can provide more complementary information to help classify the samples from the public dataset Alzheimer's Disease Neuroimaging Initiative (ADNI). In addition, the combination of modalities and different tasks are external factors that affect the performance of multimodal learning. Above all, we summrise three main problems in the early diagnosis of Alzheimer's disease: (i) unimodal vs multimodal; (ii) different combinations of modalities; (iii) classification of different tasks. In this paper, to experimentally verify these three problems, a novel and reproducible multi-classification framework for Alzheimer's disease early automatic diagnosis is proposed to evaluate and verify the above issues. The multi-classification framework contains four layers, two types of feature representation methods, and two types of models to verify these three issues. At the same time, our framework is extensible, that is, it is compatible with new modalities generated by new technologies. Following that, a series of experiments based on the ADNI-1 dataset are conducted and some possible explanations for the early diagnosis of Alzheimer's disease are obtained through multimodal learning. Experimental results show that SNP has the highest accuracy rate of 57.09% in the early diagnosis of Alzheimer's disease. In the modality combination, the addition of Single Nucleotide Polymorphism modality improves the multi-modal machine learning performance by 3% to 7%. Furthermore, we analyse and discuss the most related Region of Interest and Single Nucleotide Polymorphism features of different modalities.

Pest management is an important factor affecting agricultural and food industry products. A large number of insect species and the subtle differences bring a challenge to the accurate recognition of pests. Many studies tackle the challenge of pest recognition through deep neural networks (DNNs) and achieve significant success in terms of accuracy. However, the complex structure and a large number of parameters make DNNs difficult for end users to understand the reasons for the decision of models, which causes distrust in the classification of harmful insects and overuse of insecticides. To address the lack of explainability of DNNs, we propose an explanation generation workflow to generate concept-based explanations for pest recognition. Specifically, the concept extraction method uses a clustering algorithm to extract image segments with meaningful concepts from a portion of the training dataset. Then, concept models are trained to detect the presence of concepts in the image. Finally, the explanation generation method provides concept-based global and local explanations in the form of weighted directed graphs and concept importances, respectively. Through qualitative and quantitative analysis, the proposed workflow extracts meaningful concepts for pest recognition effectively and detects the presence of concepts in images.

With the increasing demand for cost-effective crop pest management solutions, how to achieve effective and efficient automatic pest detection has become the primary research problem. Traditional object detection methods that rely on the quality of handcrafted feature selection are hardly used in pest detection due to the difficulty of designing the features of multiple types of pests. The application of deep learning which presents outstanding performances in object detection tasks faces the following challenges in the field of pest detection. First, the detection difficulties caused by tiny-size pests and protective colouration limit the accuracy of detection. Second, pest detection requires the employment of experts to obtain the annotation of pests for training models, which is costly. Finally, the ability to run on lightweight devices is required due to the limitations of the field environment on networks and equipment. To solve these problems, this paper focuses on a lightweight tiny object detection model, training on limited supervised samples through different data augmentation methods. Different components of object detection models and data augmentation methods are analysed in different sizes of training datasets. Finally, a method based on the Yolo detection model is proposed for pest detection. This pest detection model is evaluated on a real-world aphids data set containing 6k objects. Five sets of data augmentation methods are used on seven sizes of training data sets for analysis. Then the structure of the detection neck of the Yolo model is analysed. Our experimental results show that 54.35% mAP can be achieved by the PAN module and removing the Mosaic data augmentation method for tiny object detection with one hundred samples.

Alzheimer’s disease (AD) is a neurodegenerative disease. Its condition is irreversible and ultimately fatal. Researchers have been studying approaches to support early diagnosis of Alzheimer disease and further delay the patient’s condition and improve AD patient’s quality of life. Gene expression data is a mature technology. It has many advantages such as high throughput, less-invasiveness, and affordability. It has great potential to help people diagnose Alzheimer’s disease in early stage. However, because the amount of information is too large compared to the number of samples in the Alzheimer’s database, researchers are facing “curse of dimensionality” when using gene expression data. In this work we are interested in the task of dimensionality reduction of gene expression data in Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. We investigated six dimensionality reduction algorithms: Principal component analysis, Kernel principal component analysis, Isometric feature mapping, Local linear embedding, Stacked denoising autoencoder, and Stacked sparse autoencoder. An SVM classifier is used to classify the samples in the ADNI dataset using the features obtained by dimensionality reduction. We first optimized the appropriate number of hidden layers for the two stacked autoencoders. Then we performed different degrees of dimensionality reduction with the other four algorithms and compared the classification performance of the features obtained by different algorithms with the SVM classifier. Our experimental results show that the features obtained by kernel principal component analysis and local linear embedding, Stacked denoising autoencoder, and Stacked sparse autoencoder dimensionality reduction have good AD classification performance.

The bag of visual words (BOW) model has been widely applied in the field of image recognition and image classification. However, all scale‐invariant feature transform (SIFT) features are clustered to construct the visual words which result in a substantial loss of discriminative power for the visual words. The corresponding visual phrases will further render the generated BOW histogram sparse. In this study, the authors aim to improve the classification accuracy by extracting high discriminative SIFT features and feature pairs. First, high discriminative SIFT features are extracted with the within‐ and between‐class correlation coefficients. Second, the high discriminative SIFT feature pairs are selected by using minimum spanning tree and its total cost. Next, high discriminative SIFT features and feature pairs are exploited to construct the visual word dictionary and visual phrase dictionary, respectively, which are concatenated to a joint histogram with different weights. Compared with the state‐of‐the‐art BOW‐based methods, the experimental results on Caltech 101 dataset show that the proposed method has higher classification accuracy.

The traditional [Formula: see text]-means algorithm has been widely used as a simple and efficient clustering method. However, the performance of this algorithm is highly dependent on the selection of initial cluster centers. Therefore, the method adopted for choosing initial cluster centers is extremely important. In this letter, we redefine the density of points according to the number of its neighbors, as well as the distance between points and their neighbors. In addition, we define a new distance measure that considers both Euclidean distance and density. Based on that, we propose an algorithm for selecting initial cluster centers that can dynamically adjust the weighting parameter. Furthermore, we propose a new internal clustering validation measure, the clustering validation index based on the neighbors (CVN), which can be exploited to select the optimal result among multiple clustering results. Experimental results show that the proposed algorithm outperforms existing initialization methods on real-world data sets and demonstrates the adaptability of the proposed algorithm to data sets with various characteristics.