The inland aquaculture environment is an artificial ecosystem, where the water quality is a key factor which is closely related to the economic benefits of inland aquaculture and the quality of aquatic products. Compared with marine aquaculture, inland aquaculture is normally smaller and susceptible to pollution, with poor self-purification capacity. Considering its low cost and large-scale monitoring ability, many researches have developed spectrum sensor on-board satellite platforms to allow remote monitoring of inland water surface. However, there remain many problems, such as low image resolution, poor flexible data acquisition, and anti-interference. Apart from that, the conventional forecasting model is of weak generalization ability and low accuracy. In our study, we combine unmanned aerial vehicles system (UAVs) with the wireless sensor network (WSN) to design a new ground water quality parameter and drone spectrum information acquisition approach, and to propose a novel dynamic network surgery-deep neural networks (DNS-DNNs) model based on multi-source feature fusion to forecast the distribution of dissolved oxygen (DO) and turbidity (TUB) in inland aquaculture areas. The result of using fused features, including characteristic spectrum, Gray-level co-occurrence matrix (GLCM) texture feature, and convolutional neural network (CNN) texture feature to build a model is that the characteristic spectrum+ CNN texture fusion features were the best input items for DNS-DNNs when forecasting DO, with the determination coefficient $R^2$ of the vertical set arriving at 0.8741, while the characteristic spectrum+ GLCM texture+ CNN texture fusion features were the best for TUB, with the $R^2$ reaching 0.8531. Compared with a variety of conventional models, our model had a better performance in the inversion of DO and TUB, and there was a strong correlation between predicted and real values: $R^2$ reached 0.8042 and 0.8346, whereas the root mean square error (RMSE) were only 0.1907 and 0.1794, separately. Our study provides a new insight about using remote sensing to rapidly monitor water quality in inland aquaculture regions.

Deep learning has been widely used in various sensor detection fields to realize intelligent detection tasks. However, there are always restrictions on how these functions can be integrated into embedded devices to complete mobile and convenient sensing detection tasks. Therefore, the MobileNet_SSD, A lightweight model after simplification and optimization, is used in this paper to design a portable visual sensor system for the detection of spray droplet deposition in drone applications. The sensing system includes a droplet deposition image loop acquisition device and a supporting host computer interactive platform. It introduces the Sarsa algorithm to establish an adaptive model of light intensity to enhance the robustness and combines Deep Neural Networks and image processing technology to achieve rapid loop detection of droplet deposition parameters on the mobile terminal. Experimental results indicate that the proposed sensor can adapt to light changes in complex environments, and accurately measure the deposition parameters of different droplet density images. Furthermore, it is of great significance to detect the quality of drone sprays, master the rules of droplet deposition, and understand the effects of pesticide spraying.

Vision-based adaptive variable rate spraying approach for unmanned aerial vehicles

The rapid development of vision sensor based on artificial intelligence (AI) is reforming industries and making our world smarter. Among these trends, it is of great significance to adapt AI technologies into the intelligent agricultural management. In smart agricultural aviation spraying, the droplets’ distribution and deposition are important indexes for estimating effectiveness in plant protection process. However, conventional approaches are problematic, they lack adaptivity to environmental changes, and consumes non-reusable test materials. One example is that the machine vision algorithms they employ can’t guarantee that the division of adhesive droplets thereby disabling the accurate measurement of critical parameters. To alleviate these problems, we put forward an intelligent visual droplet detection node which can adapt to the environment illumination change. Then, we propose a modified marker controllable watershed segmentation algorithm to segment those adhesive droplets, and calculate their characteristic parameters on the basis of the segmentation results, including number, coverage, coverage density, etc. Finally, we use the intelligent node to detect droplets, and then expound the situation that the droplet region is effectively segmented and marked. The intelligent node has better adaptability and robustness even under the condition of illumination changing. The large-scale distributed detection result indicates that our approach has good consistency with the non-recyclable water-sensitive paper approach. Our approach provides an intelligent and environmental friendly way of tests for spraying techniques, especially for plant protection with Unmanned Aerial Vehicles.

Adaptive target spray system based on machine vision for plant protection UAV

With computer vision booming ahead on the high-way, image classification and object detection have greatly improved, which are able to accurately classify large datasets even better than human beings. There are many fields in which computer vision gets involved, including terrestrial and aerial mapping of natural resources, crop monitoring, precision agriculture, quality control and classification of processing lines, and process automation. The application of computer vision in agriculture has increased considerably in recent years. However, traditional computer vision methods require complex preprocessing and algorithms, which is time-consuming and labor-intensive. In particular, the effectiveness of this method relies mainly on the accuracy of the artificial design features and the learning algorithm. In this paper, we propose a object detection algorithm based on Faster-RCNN to identify the stage of the cotton leaf(a period that cotton leaves grow). The algorithm consists of three main parts: convolutional neural network based on ResNet101, the extraction of candidate regions of the target based on RPNs, and the identification of detection object. Based on many experiments, the algorithm in this paper is able to identify the cotton leaf stage, and the precision surpasses 80%.