Smart agricultural technology最新文献

Progress in non-destructive methods to detect the characteristics of fruits is a new and attractive process for researchers and specialists in this field. On the other hand, these researchers move toward identifying their impacts on their surroundings in line with diagnostic efficiency. One of these essential impacts is the environmental impact of the non-destructive detection process of fruits. Navel oranges are one of the most popular and widely consumed fruits, whose maturity indices such as soluble solids contents (SSC) values and acidity are considered as parameters in determining the quality of this product. This study used the hyperspectral method in the vis-NIR range to evaluate and measure navel oranges' SSC and acidity values. In the following, by applying the life cycle assessment method, the environmental impacts of measuring and evaluating these two parameters of the characteristics of navel oranges were investigated. The Impact2002+ method was used to evaluate the impact of the life cycle list. Based on the findings, the environmental impacts of SSC measurement are about 40, 42, 20, and 18 % higher than those of the environmental impacts of pH measurement from the point of view of endpoint impacts for Human Health, Ecosystem quality, climate change, and resources, respectively. The random forest modeling results showed a suitable and acceptable correlation and relationship (over 90 %) between the wavelengths selected from the feature selection stage and environmental impacts.

Accurate and early detection of insect pests plays an important role in crop protection and pest management in agriculture, especially in orchards. This paper is focused on evaluating and improving the performance of insect detection algorithms by adopting an ensemble approach of artificial neural networks. A set of advanced object detection models including YOLOv8, Faster R-CNN, RetinaNet, SSD, and FCOS were selected, and the models were trained and evaluated on a common dataset representing digital images of different insect species pests. Two classes were considered represented by quite similar invasive stink bugs, Halyomorpha Halys and Nezara Viridula. These architectures were optimized to identify significant peculiarities and variations between reference insects, including size, shape, and color. Each model has been implemented and optimized to achieve the best possible performance before integrating into an ensemble system. By integrating the predictions of these models through a weighted ensemble mechanism that leverages the F1 Score of each model, a more performant global system was developed capable of detecting insect pests with improved performance over individual models. This significant improvement in insect detection highlights the potential of the proposed ensemble system in efficient and rapid insect pest identification, ultimately providing valuable opportunities for implementing crop monitoring technologies. The research highlights the importance of implementing and developing deep-learning technologies for solving specific challenges in agriculture and brings innovative ways of strategic pest management for sustainable agricultural practices.

The identification of individual trees is an important research topic in forestry, remote sensing and computer vision. It represents a tool for effectively and efficiently managing and maintaining forests and orchards. However, this task is not as simple as it seems; tree detection and counting can be time consuming, cost-prohibitive and accuracy-limited, especially if performed manually on a large scale.The availability of very high-resolution UAV imagery with remote sensing can make the counting process easier, faster and more precise. With the development of technology, this process can be made more automated by using intelligent algorithms such as CNN.

This work presents an OBIA-CNN (Object Based Image Analysis-Convolution Neural Network) approach that combines CNNs with OBIA to automatically detect and count olive trees from Phantom4 advanced drone imagery. Initially, The CNN-based classifier was created, trained, validated, and applied to generate the Olive trees probability maps on the ortho-photo. The post-classification refinement based on OBIA was then conducted. A super-pixel segmentation and the Excess Green index were performed and a detailed accuracy analysis has been carried out to establish the suitability of the proposed method.

The application to a RGB ortho-mosaic of an olive grove, in the east region of Morocco was successful using a manually elaborated training dataset of 4500 images of 24×24 pixels. Finally, the CNN detected and counted 2934 olive trees on the ortho-photo, achieving an overall accuracy of 97 % and 99 % after the OBIA refinement. The results of the proposed OBIA-CNN method were also compared with the classification results of using the Template matching technique, CNN method alone, and OBIA analysis alone to evaluate the performance of the approach. Our findings suggest the use of very high resolution images with object-based deep learning is promising for automatic detection and counting of olive trees to support the accurate and sustainable agricultural monitoring.

Animal behavior monitoring is an important tool for animal production. This behavior monitoring strategy can indicate the well-being and health of animals, which can lead to better productive performance. This study aimed to assess the most effective accelerometer attachment position (on either the halter or a neck collar) and data transmission time intervals (ranging from 6 to 600 s) for predicting behavioral patterns, including water and food intake frequencies, as well as other activities in young beef cattle bulls within a feedlot system. A range of machine learning algorithms were applied to satisfy the aims of the study, including the random forest, support vector machine, multilayer perceptron, and naive Bayes classifier algorithms. All studied models produced high performance metrics (above 0.90) when using both attachment positions, except for the models built using the naive Bayes classifier. Therefore, coupling accelerometers with collars is a more viable alternative for use on animals, as doing so is easier than applying accelerometers to halters. Utilizing a dataset with more observations (i.e., shorter time intervals) did not result in considerable improvements in the performance metrics of the trained models. Therefore, using datasets with fewer observations is more advantageous, as it can lead to decreased computational and temporal demands for model training, in addition to saving the battery of the device considered in this study.

Radiation is a crucial factor in the field of agrivoltaics in greenhouses. Depending on the type of photovoltaics integrated into greenhouses, the effect on radiation varies through the phenomenon of shading. Shading in greenhouses can be either beneficial or detrimental, making its analysis imperative. In this study, the improvements and modifications made in an algorithm capable of calculating the shading from photovoltaic units installed in greenhouses are presented. A key modification to the algorithm is the calculation of shading from semi-transparent photovoltaic modules, in contrast to its original form, in which photovoltaic modules were considered opaque. The algorithm was validated using radiation data from pyranometers within a greenhouse. The coefficients used were Pearson's r correlation coefficient and the Coefficient of Variation. The correlation coefficients for times without shading effect approached the values for the case without photovoltaics installed on the roof. Simultaneously, based on the coefficients of variation, the uniformity of radiation within the shade was validated, thereby confirming its existence. Finally, the effect of semi-transparent photovoltaic units on Global Horizontal Irradiance and Photosynthetically Active Radiation was studied, with the reduction for the former approaching 52 % and about 60 % for the latter. These changes in radiation can be either beneficial or not, depending on the type of crop and the needs of the greenhouse, such as cooling.

Mastitis is a significant disease on dairy farms and can have serious negative animal performance and economic consequences if not controlled. While clinical mastitis is often easily identified due to visibly abnormal milk, subclinical mastitis presents a more insidious challenge. Somatic cell count (SCC) is commonly used to monitor and detect subclinical mastitis, however, SCC is not available at a high sampling frequency rate at the cow level on most farms due to the manual effort involved in collecting it. With the rise of precision dairy farming technologies such as milk meters, however, there is increasing interest in using data-driven approaches (especially approaches using machine learning) for detecting subclinical mastitis based on indicators more easily collected by modern sensors. In this article we introduce milk flow profiles, a new, easy-to-collect data type that can replace more difficult-to-collect data sources (e.g., those that require laboratory tests or manual measurements) in precision dairy farming. The results of our experiments demonstrate that milk flow profiles, combined with other easily accessible milking machine data, can be employed to train machine learning models that accurately detect subclinical mastitis (as evidenced by high SCC measurements), with an AUC of 0.793. Moreover, these models perform better than models trained using features from milk characteristic data that are expensive to collect and are only collected at low frequency on commercial farms. Our experiments used data from 16 weeks of milking events from 285 cows on Irish farms, and their results demonstrate the value of milk flow profiles as an easily accessible and valuable data source for precision dairy farming applications.

Technological advancements have revolutionized livestock farming, notably in health monitoring. Traditional methods, which have been criticized for subjectivity and treatment delays, can be replaced with efficient health monitoring systems, thereby reducing costs and workload. Implementing cow behavior recognition allows for effective dairy cow health monitoring. In this research, we propose an integrated system using inertial measurement unit (IMU) devices and machine learning techniques for dairy cow behavior recognition. Six main dairy cow behaviors were studied: lying, standing, walking, drinking, feeding, and ruminating. All behavior types were manually labeled into the IMU data by reviewing the recorded footage. The labeled IMU data underwent four processing steps: selecting different window sizes, feature extraction, feature selection, and normalization. These processed data were then used to build the behavior recognition model. Various model structures, including SVM, Random Forest, and XGBoost, were tested. The top-performing model, XGBoost, with its proposed 58 features achieved an F1-score of 0.87, with specific scores of 0.93 for lying, 0.85 for walking, 0.94 for ruminating, 0.89 for feeding, 0.86 for standing, 0.93 for drinking, and 0.59 for other activities. During our online testing, we observed similar patterns for each healthy cow. The cumulative time spent on each behavior also matched the statistics from previous surveys. Additionally, our backend optimization approach resulted in a final overall percentage error of 1.55 % per day during online testing. In conclusion, our study presents an IMU-based system capable of accurately recognizing dairy cow behavior. Feature design and appropriate models are proposed herein. A functional optimization method was introduced indicating that our system has the potential with applications for estrus detection and other reproductive management practices in the dairy industry.

Observing poultry activity is crucial for assessing their health status; however, the inspection process is often time-consuming and labor-intensive, particularly in cases involving large numbers of chickens. Inexperienced breeders may also misjudge their activity levels, potentially missing opportunities for prevention and treatment. This study integrates traditional video surveillance with an advanced monitoring system to identify various broiler behaviors in a breeding environment. A two-stage deep learning approach is employed: in the first stage, the broilers are detected, and in the second stage, five key body points (head, abdomen, two legs, and tail) are identified. A skeleton-based model is then developed centered around the abdomen, with six angles calculated using trigonometric methods. These angles are analyzed by a long short-term memory network to estimate behaviors such as “Standing”, “Walking”, “Resting”, “Eating”, “Preening”, and “Flapping”, selecting the behavior with the highest probability. Dual-layer fuzzy logic inference systems were used to evaluate the proportion of time broilers spent in static versus dynamic states, providing a robust determination of their activity levels. Validated in a mixed-sex breeding environment, the proposed system achieved accuracies of at least 85.2% for identifying broiler type, 79.2% for identifying body parts, and 50.8% for identifying behaviors. The activity level evaluation results were consistent with those conducted by experienced poultry experts.

The implementation of a machine-vision system for real-time precision weed management is a crucial step towards the development of smart spraying robotic vehicles. The intelligent machine-vision system, constructed using deep learning object detection models, has the potential to accurately detect weeds and crops from images. Both data-centric and model-centric approaches of deep learning model development offer advantages depending on environment and non-environment factors. To assess the performance of weed detection in real-field conditions for eight crop species, the Yolov8, Yolov9, and customized Yolov9 deep learning models were trained and evaluated using RGB images from four locations (Casselton, Fargo, and Carrington) over a two-year period in the Great Plains region of the U.S. The experiment encompassed eight crop species—dry bean, canola, corn, field pea, flax, lentil, soybean, and sugar beet—and five weed species—horseweed, kochia, redroot pigweed, common ragweed, and water hemp. Six Yolov8 and eight Yolov9 model variants were trained using annotated weed and crop images gathered from four different sites including combined dataset from four sites. The Yolov8 and Yolov9 models’ performance in weed detection were assessed based on mean average precision (mAP50) metrics for five datasets, eight crop species, and five weed species. The results of the weed and crop detection evaluation showed high mAP50 values of 86.2 %. The mAP50 values for individual weed and crop species detection ranged from 80.8 % to 98 %. The results demonstrated that the performance of the model varied by model type (model-centric), location due to environment (data-centric), data size (data-centric), data quality (data-centric), and object size in the image (data-centric). Nevertheless, the Yolov9 customized lightweight model has the potential to play a significant role in building a real time machine-vision-based precision weed management system.

This paper presents the development of a robotic sprayer that combines artificial intelligence with robotics for optimal spray application on citrus nursery plants grown in an indoor environment. The robotic platform is integrated with an embedded firmware of MobileNetV2 model to identify and classify the plant samples with a classification accuracy of 100 % which is used to dispense variable rate spraying of pesticide based on the health status of the plant foliage. The disease detection model was developed through the edge impulse platform and deployed on Raspberry Pi 4. The robot navigates through an array of plants, stops beside each plant, and captures an image of the citrus plants. It feeds the image into the deployed embedded model to generate a disease inference that informs the variable rate application of spray during real-time actuation. To test the spraying performance of the prototype within the growing environment, water sensitive cards were placed in each plant's canopy. After spraying, the samples of water sensitive cards were collected and quantified using a smart spray app to determine the classification accuracy as well as the extent of spray coverage on the citrus samples. The robot spray coverage results show an average spray coverage of 87 % on lemon foliage when compared with 67 % for navel orange, during the spray performance test of the robot.