Detecting tea tree pests in complex backgrounds using a hybrid architecture guided by transformers and multi-scale attention mechanism.

BACKGROUND: Tea pests pose a significant threat to tea leaf yield and quality, necessitating fast and accurate detection methods to improve pest control efficiency and reduce economic losses for tea farmers. However, in real tea gardens, some tea pests are small in size and easily camouflaged by complex backgrounds, making it challenging for farmers to promptly and accurately identify them. RESULTS: To address this issue, we propose a real-time detection method based on TP-YOLOX for monitoring tea pests in complex backgrounds. Our approach incorporates the CSBLayer module, which combines convolution and multi-head self-attention mechanisms, to capture global contextual information from images and expand the network's perception field. Additionally, we integrate an efficient multi-scale attention module to enhance the model's ability to perceive fine details in small targets. To expedite model convergence and improve the precision of target localization, we employ the SIOU loss function as the bounding box regression function. Experimental results demonstrate that TP-YOLOX achieves a significant performance improvement with a relatively small additional computational cost (0.98 floating-point operations), resulting in a 4.50% increase in mean average precision (mAP) compared to the original YOLOX-s. When compared with existing object detection algorithms, TP-YOLOX outperforms them in terms of mAP performance. Moreover, the proposed method achieves a frame rate of 82.66 frames per second, meeting real-time requirements. CONCLUSION: TP-YOLOX emerges as a proficient solution, capable of accurately and swiftly identifying tea pests amidst the complex backgrounds of tea gardens. This contribution not only offers valuable insights for tea pest monitoring but also serves as a reference for achieving precise pest control. © 2023 Society of Chemical Industry.

A Real-Time Smooth Weighted Data Fusion Algorithm for Greenhouse Sensing Based on Wireless Sensor Networks.

Wireless sensor networks are widely used to acquire environmental parameters to support agricultural production. However, data variation and noise caused by actuators often produce complex measurement conditions. These factors can lead to nonconformity in reporting samples from different nodes and cause errors when making a final decision. Data fusion is well suited to reduce the influence of actuator-based noise and improve automation accuracy. A key step is to identify the sensor nodes disturbed by actuator noise and reduce their degree of participation in the data fusion results. A smoothing value is introduced and a searching method based on Prim's algorithm is designed to help obtain stable sensing data. A voting mechanism with dynamic weights is then proposed to obtain the data fusion result. The dynamic weighting process can sharply reduce the influence of actuator noise in data fusion and gradually condition the data to normal levels over time. To shorten the data fusion time in large networks, an acceleration method with prediction is also presented to reduce the data collection time. A real-time system is implemented on STMicroelectronics STM32F103 and NORDIC nRF24L01 platforms and the experimental results verify the improvement provided by these new algorithms.

Adaptive Energy-Efficient Target Detection Based on Mobile Wireless Sensor Networks.

Target detection is a widely used application for area surveillance, elder care, and fire alarms; its purpose is to find a particular object or event in a region of interest. Usually, fixed observing stations or static sensor nodes are arranged uniformly in the field. However, each part of the field has a different probability of being intruded upon; if an object suddenly enters an area with few guardian devices, a loss of detection will occur, and the stations in the safe areas will waste their energy for a long time without any discovery. Thus, mobile wireless sensor networks may benefit from adaptation and pertinence in detection. Sensor nodes equipped with wheels are able to move towards the risk area via an adaptive learning procedure based on Bayesian networks. Furthermore, a clustering algorithm based on k-means++ and an energy control mechanism is used to reduce the energy consumption of nodes. The extended Kalman filter and a voting data fusion method are employed to raise the localization accuracy of the target. The simulation and experimental results indicate that this new system with adaptive energy-efficient methods is able to achieve better performance than the traditional ones.

Blind RSSD-Based Indoor Localization with Confidence Calibration and Energy Control.

Indoor localization based on wireless sensor networks (WSNs) is an important field of research with numerous applications, such as elderly care, miner security, and smart buildings. In this paper, we present a localization method based on the received signal strength difference (RSSD) to determine a target on a map with unknown transmission information. To increase the accuracy of localization, we propose a confidence value for each anchor node to indicate its credibility for participating in the estimation. An automatic calibration device is designed to help acquire the values. The acceleration sensor and unscented Kalman filter (UKF) are also introduced to reduce the influence of measuring noise in the application. Energy control is another key point in WSN systems and may prolong the lifetime of the system. Thus, a quadtree structure is constructed to describe the region correlation between neighboring areas, and the unnecessary anchor nodes can be detected and set to sleep to save energy. The localization system is implemented on real-time Texas Instruments CC2430 and CC2431 embedded platforms, and the experimental results indicate that these mechanisms achieve a high accuracy and low energy cost.

Energy-Efficient Control with Harvesting Predictions for Solar-Powered Wireless Sensor Networks.

Wireless sensor networks equipped with rechargeable batteries are useful for outdoor environmental monitoring. However, the severe energy constraints of the sensor nodes present major challenges for long-term applications. To achieve sustainability, solar cells can be used to acquire energy from the environment. Unfortunately, the energy supplied by the harvesting system is generally intermittent and considerably influenced by the weather. To improve the energy efficiency and extend the lifetime of the networks, we propose algorithms for harvested energy prediction using environmental shadow detection. Thus, the sensor nodes can adjust their scheduling plans accordingly to best suit their energy production and residual battery levels. Furthermore, we introduce clustering and routing selection methods to optimize the data transmission, and a Bayesian network is used for warning notifications of bottlenecks along the path. The entire system is implemented on a real-time Texas Instruments CC2530 embedded platform, and the experimental results indicate that these mechanisms sustain the networks' activities in an uninterrupted and efficient manner.