The appearance of pest insects can lead to a loss in yield if farmers do not respond in a timely manner to suppress their spread. Occurrences and numbers of insects can be monitored through insect traps, which include their permanent touring and checking of their condition. Another more efficient way is to set up sensor devices with a camera at the traps that will photograph the traps and forward the images to the Internet, where the pest insect’s appearance will be predicted by image analysis. Weather conditions, temperature and relative humidity are the parameters that affect the appearance of some pests, such as Helicoverpa armigera. This paper presents a model of machine learning that can predict the appearance of insects during a season on a daily basis, taking into account the air temperature and relative humidity. Several machine learning algorithms for classification were applied and their accuracy for the prediction of insect occurrence was presented (up to 76.5%). Since the data used for testing were given in chronological order according to the days when the measurement was performed, the existing model was expanded to take into account the periods of three and five days. The extended method showed better accuracy of prediction and a lower percentage of false detections. In the case of a period of five days, the accuracy of the affected detections was 86.3%, while the percentage of false detections was 11%. The proposed model of machine learning can help farmers to detect the occurrence of pests and save the time and resources needed to check the fields.
While past research has emphasized the importance of late blight infection detection and classification, anticipating the potato late blight infection is crucial from the economic point of view as it helps to significantly reduce the production cost. Furthermore, it is necessary to minimize the exposure of potatoes to harmful chemicals and pesticides due to their potential adverse effects on the human immune system. Our work is based on the precise classification of late blight infections in potatoes in European countries using real-time data from 1980 to 2000. To predict the potato late blight outbreak, we incorporated several hybrid machine learning models, as well as a unique combination of stacking classifier and logistic regression, achieving the highest prediction accuracy of 87.22%. Further enhancements of these models and the use of new data sources may lead to a higher late blight prediction accuracy and, consequently, a higher efficiency in managing potatoes’ health.
The widespread use of IoT devices has led to the generation of a huge amount of data and driven the need for analytical solutions in many areas of human activities, such as the field of smart agriculture. Continuous monitoring of crop growth stages enables timely interventions, such as control of weeds and plant diseases, as well as pest control, ensuring optimal development. Decision-making systems in smart agriculture involve image analysis with the potential to increase productivity, efficiency and sustainability. By applying Convolutional Neural Networks (CNNs), state recognition and classification can be performed based on images from specific locations. Thus, we have developed a solution for early problem detection and resource management optimization. The main concept of the proposed solution relies on a direct connection between Cloud and Edge devices, which is achieved through Fog computing. The goal of our work is creation of a deep learning model for image classification that can be optimized and adapted for implementation on devices with limited hardware resources at the level of Fog computing. This could increase the importance of image processing in the reduction of agricultural operating costs and manual labor. As a result of the off-load data processing at Edge and Fog devices, the system responsiveness can be improved, the costs associated with data transmission and storage can be reduced, and the overall system reliability and security can be increased. The proposed solution can choose classification algorithms to find a trade-off between size and accuracy of the model optimized for devices with limited hardware resources. After testing our model for tomato disease classification compiled for execution on FPGA, it was found that the decrease in test accuracy is as small as 0.83% (from 96.29% to 95.46%).
Colonies of the European honeybee, the most important pollinator insects, are subjected to many negative impacts of modern society, in form of pollution, pests, invading species and colony collapse disorder (CCD). The newly emerging technology of the Internet of Things (IoT), enables remote monitoring using wireless sensors inside beehives. In this paper, we present the model of the system for remote temperature monitoring on a various points inside the beehive colony using IoT devices. Measured data is transmitted to a remote server where received data streams are processed in real-time using Complex Event Processing (CEP) which enables detection of critical events and report it to the beekeeper. These data streams are compared with reference temperature patterns using machine learning algorithms, which give computers the ability to learn to detect events without being explicitly programmed. This system can significantly reduce the beekeeper reaction time and increase chances for a beehive colony overcoming certain types of anomalies with human intervention.
Necessity for seasonal monitoring of economically most important pests in different crops, increase input costs of their surveillance. In maize fields, Western corn rootworm - WCR (Diabrotica virgifera sp. virgifera) is economically the most important species and presents a limiting factor for production of maize in continuous cultivation. Its occurrence is usually monitored with pheromone traps. However, as previously mentioned they are time and money consuming due to constant need for field inspections. Since in research projects, finances predefined for weekly inspection of traps are most often non-eligible, there is a need for developing a novel approach for pest monitoring. The use of IT technologies along with commercially available pheromone traps could provide precise information about the situation in traps without frequent field inspections. Also, they are easy to maintain, manipulate and require minimum costs. This work aimed to assess the potential use and the precision of a sensor device with camera, in monitoring the WCR flight on pheromone traps. Sensor device equipped with small camera can capture images of a pheromone trap sticky base and transfer them to a remote server for review, storage and analysis. The main idea in this paper is to present a system that uses a method based on analysis of the image variations of the pheromone pest traps, performed on devices placed in their vicinity which means that sending every image to the server is avoided. In this way, information about the variations in pheromone traps can be found in one location without unnecessarily sending the same images to the server. The obtained results indicate that the proposed method for monitoring the variations of the number of caught specimens on sticky surfaces of pheromone traps, based on the variations of the dark surface on the images, can be a reliable tool in further work.
We present a deep learning model for the recognition of speech commands in the English language. The dataset is based on the Google Speech Commands Dataset by Warden P., version 0.01, and it consists of ten distinct commands (“left”, “right”, “go”, “stop”, “up”, “down”, “on”, “off”, “yes”, and “no”) along with additional “silence” and “unknown” classes. The dataset is split in a speaker-independent manner, with 70% of speakers assigned to the training set and 15% to the test set and validation set. All audio clips are sampled at 16 kHz, with a total of 46 146 clips. Audio files are converted into Mel spectrogram representations, which are then used as input to a deep learning model composed of a four-layer convolutional neural network followed by two fully connected layers. The model employs Rectified Linear Unit (ReLU) activation, the Adam optimizer, and dropout regularization to improve generalization. The achieved testing accuracy is 96.05%. Micro- and macro-averaged precision, recall, and F1-score of 95% are reported to reflect class-wise performance, and a confusion matrix is also provided. The proposed model has been deployed on a Raspberry Pi 5 as a Fog computing device for real-time speech recognition applications.