Artificial Intelligence in Agriculture最新文献

Macronutrient deficiency inhibits the growth and development of chili plants. One of the non-destructive methods that plays a role in processing plant image data based on specific characteristics is computer vision. This study uses 5166 image data after augmentation process for six plant health conditions. But the analysis of one feature cannot represent plant health condition. Therefore, a careful combination of features is required. This study combines three types of features with HSV and RGB for color, GLCM and LBP for texture, and Hu moments and centroid distance for shapes. Each feature and its combination are trained and tested using the same MLP architecture. The combination of RGB, GLCM, Hu moments, and Distance of centroid features results the best performance. In addition, this study compares the MLP architecture used with previous studies such as SVM, Random Forest Technique, Naive Bayes, and CNN. CNN produced the best performance, followed by SVM and MLP, with accuracy reaching 97.76%, 90.55% and 89.70%, respectively. Although MLP has lower accuracy than CNN, the model for identifying plant health conditions has a reasonably good success rate to be applied in a simple agricultural environment.

The expression of quantitative traits of a line of a crop depends on its genetics, the environment where it is sown and the interaction between the genetic information and the environment known as GxE. Thus to maximize food production, new varieties are developed by selecting superior lines of seeds suitable for a specific environment. Genomic selection is a computational technique for developing a new variety that uses whole genome molecular markers to identify top lines of a crop. A large number of statistical and machine learning models are employed for single environment trials, where it is assumed that the environment does not have any effect on the quantitative traits. However, it is essential to consider both genomic and environmental data to develop a new variety, as these strong assumptions may lead to failing to select top lines for an environment. Here we devised three novel deep learning frameworks incorporating GxE within the deep learning model and predicted line-specific yield for an environment. In the process, we also developed a new technique for identifying environment-specific markers that can be useful in many applications of environment-specific genomic selection. The result demonstrates that our best framework obtains 1.75 to 1.95 times better correlation coefficients than other deep learning models that incorporate environmental data depending on the test scenario. Furthermore, the feature importance analysis shows that environmental information, followed by genomic information, is the driving factor in predicting environment-specific yield for a line. We also demonstrate a way to extend our framework for new data types, such as text or soil data. The extended model also shows the potential to be useful in genomic selection.

The spatial distribution of laying hens in cage-free houses is an indicator of flock's health and welfare. While larger space allows chickens to perform more natural behaviors such as dustbathing, foraging, and perching in cage-free houses, an inherent challenge is evaluating chickens' locomotion and spatial distribution (e.g., real-time birds' number on perches or in nesting boxes). Manual inspection of hen's spatial distribution requires closer observation, which is labor intensive, time consuming, subject to human errors, and stress causing on birds. Therefore, an automated monitoring system is required to track the spatial distribution of hens for early detection of animal welfare and health concerns. In this study, a non–intrusive machine vision method was developed to monitor hens' spatial distribution automatically. An improved You Only Look Once version 5 (YOLOv5) method was developed and trained to test hens' distribution in research cage-free facilities (e.g., 200 hens per house). The spatial distribution of hens the system monitored includes perch zone, feeding zone, drinking zone, and nesting zone. The dataset contains a whole growth period of chickens from day 1 to day 252. About 3000 images were extracted randomly from recorded videos for model training, validation, and testing. About 2400 images were used for training and 600 images for testing, respectively. Results show that the accuracy of the new model were 87–94% for tracking distribution in different zones for different ages of hens/pullets. Birds' age affected the performance of the model as younger birds had smaller body size and were hard to be detected due to blackness or occultation by equipment. The performance of the model was 0.891 and 0.942 for baby chicks (≤10 days old) and older birds (> 10 days) in detecting perching behaviors; 0.874 and 0.932 in detecting feeding/drinking behaviors. Miss detection happened when the flock density was high (>18 birds/m²) and chicken body was occluded by other facilities (e.g., nest boxes, feeders, and perches). Further studies such as chicken behavior identification works in commercial housing system should be combined with the model to reach an automatic detection system.

The generation of food production that meets the rising demand for food and ecosystem security is a big challenge. With the development of Artificial Intelligence (AI) models, there is a growing need to use them to achieve sustainable agriculture. The continuous enhancement of AI in agriculture, researchers have proposed many models in agriculture functions such as prediction,weed control, resource management, advance care of crops, and so on. This article evaluates on a systematic review of AI models in agriculture functions. It also reviews how AI models are used in identified sustainable objectives. Through this extensive review, this paper discusses considerations and limitations for building the next generation of sustainable agriculture using AI.

Fruit is a key crop in worldwide agriculture feeding millions of people. The standard supply chain of fruit products involves quality checks to guarantee freshness, taste, and, most of all, safety. An important factor that determines fruit quality is its stage of ripening. This is usually manually classified by field experts, making it a labor-intensive and error-prone process. Thus, there is an arising need for automation in fruit ripeness classification. Many automatic methods have been proposed that employ a variety of feature descriptors for the food item to be graded. Machine learning and deep learning techniques dominate the top-performing methods. Furthermore, deep learning can operate on raw data and thus relieve the users from having to compute complex engineered features, which are often crop-specific. In this survey, we review the latest methods proposed in the literature to automatize fruit ripeness classification, highlighting the most common feature descriptors they operate on.

According to the industry, the value of wood logs is heavily influenced by their internal structure, particularly the distribution of knots within the trees. Nowadays, CT scanners combined with classical computer vision approach are the most common tool for obtaining reliable and accurate images of the interior structure of trees. Knowing where the tree semantic features, especially knots, contours and centers are within a tree could improve the efficiency of the overall tree industry by minimizing waste and enhancing the quality of wood-log by-products. However, this requires to automatically process the CT-scanner images so as to extract the different elements such as tree centerline, knot localization and log contour, in a robust and efficient manner. In this paper, we propose an effective methodology based on deep learning for performing these different tasks by processing CT-scanner images with deep convolutional neural networks. To meet this objective, three end-to-end trainable pipelines are proposed. The first pipeline is focused on centers detection using CNNs architecture with a regression head, the second and the third one address contour estimation and knot detection as a binary segmentation task based on an Encoder-Decoder architecture. The different architectures are tested on several tree species. With these experiments, we demonstrate that our approaches can be used to extract the different elements of trees in a precise manner while preserving good performances of robustness. The main objective was to demonstrate that methods based on deep learning might be used and have a relevant potential for segmentation and regression on CT-scans of tree trunks.

The maturity state of durian fruit is a key indicator of quality before trading. This research aims to improve the near-infrared (NIR) model for classifying the maturity stage of durian fruit using a completely non-destructive measurement. Both NIR spectrometers were investigated: the short wavelength NIR (SWNIR) ranging from 450 to 1000 nm and long wavelength NIR (LWNIR) ranging from 860 to 1750 nm. The samples collected for experimentation consisted of four stages: immaturity, prematurity, maturity, and ripe. Each fruit was scanned at the rind position on the main fertile lobe (header, middle, and tail) and stem. The classification models were developed using three supervised machine learning algorithms: linear discriminant analysis (LDA), support vector machine (SVM), and K-Nearest neighbours (KNN). The analysis results revealed that the use of durian rind spectra only obtained between 83.15% and 88.04% accuracy for the LWNIR spectrometer, while the SWNIR spectrometer provided 64.73 to 93.77% accuracy. The performance of model increases when developing with combination between rind and stem spectra. The LDA model developed using a combination of rind and stem spectra provided the greatest efficiency, exhibiting 97.28% and 100% accuracy for LWNIR and SWNIR spectrometers, respectively. The LDA model is therefore recommended for obtaining spectra from smoothing moving average (MA) + baseline of rind position and when used in combination with the MA + standard normal variance (SNV) of stem spectra. The NIR spectroscopy indicated high potential for non-destructive estimation of the durian maturity stage. This process could be used for quality control in the durian export industry to solve the problem of unripe durian being mixed with ripe fruit.

African swine fever (ASF) is a contagious and lethal hemorrhagic disease with a high case fatality rate. Since 2007, ASF has been spreading into many countries, especially in Europe and Asia. Given that there is no effective vaccine and treatment to deal with ASF, prevention is an important way for a country to avoid the effects of the virus. Australia is currently ASF-free but the disease has been reported in many neighboring countries, such as Indonesia, Timor-Leste, and Papua New Guinea. Therefore, it is necessary for Australia to maintain hyper-vigilance to prevent the ASF introduction. In this paper, we propose the use of fuzzy concepts to establish a fuzzy risk assessment model to predict the ASF introduction risk in Australia. From the analysis, the international passengers (IP) and international import trade (IIT) are concluded as the two main ASF introduction factors based on transmission features and past research. From the established fuzzy risk assessment model based on the analysis of the 2019 and 2020 data, the risks of ASF introduction into Australia are considered to be low. The model further deduced that the Asian region was the major source of potential risks. Finally, in order to validate the effectiveness of the established fuzzy risk assessment model, the qualitative data from the Department for Environment, Food & Rural Affairs of the United Kingdom was used. From the validation results, it has shown that the results were consistent when the same data is adopted, and thus proved that the functionality of the established fuzzy risk assessment model for assessing the risk in Australia.

In recent years there is a growing importance in using machine learning techniques to improve procedures in precision agriculture: in this work we perform a study on models capable of predicting oenological parameters from hyperspectral images of wine grape berries, a specially relevant topic to boost production tasks for winemakers. Specifically, we explore the capabilities of a novel technique mostly used for visualization, t-Distributed Stochastic Neighbor Embedding (t-SNE), for reducing the dimensionality of the highly complex hyperspectral data and compare its performance with Principal Component Analysis (PCA) method, which despite the introduction of many nonlinear dimensionality reduction techniques over the years, had achieved the best results for real-world data across several studies in literature. Additionally we explore the potential of Kernel t-SNE, an extension to the t-SNE method that allows for the usage of the technique in streaming data or online scenarios. Our results show that, in a direct comparison, t-SNE achieves better metrics than PCA for most of the data sets in this work and that the regressor (Support Vector Regression, SVR) performs better with the t-SNE reduced features as inputs, accomplishing better predictions with lower error rates. Comparing the results with current literature, our shallow learning model paired with t-SNE achieves either better or on par results than those reported, even competing with more advanced models that use deep learning techniques, which should propel the introduction of t-SNE in more studies that require dimensionality reduction.