近年来，随着人类不断利用自然和改造自然，生态环境问题愈演愈烈。我国在推动国民经济合理健康发展的同时，也更加注重生态文明建设。湿地作为一种重要的生态环境资源，在生态多样性、涵养水源、防止土地沙漠化、减少温室效应等方面有着极其重要的作用。因此，对湿地的监测和保护至关重要，构建湿地生态环境监测系统是保护湿地资源重要的举措之一。传统的人工监测方式成本高，效率低下，已无法满足当代平台化、信息化、智能化湿地生态环境监测的需求。随着遥感技术、无线传感器技术、人工智能等技术不断被引入到湿地生态环境监测中去，逐渐取代了早期的人工监测手段，提升了湿地生态环境监测系统的效率及可靠性。本研究基于无线传感器技术和人工智能技术，在九里湖国家湿地公园构建了湿地生态环境监测系统（以下简称系统）。

随着九里湖生态环境的改善，形成了丰富的湿地生物群落，因此对研究区内生物多样性的保护至关重要。本研究在系统中使用改进的目标监测算法(You Only Look Once V5， YOLOV５)来开发基于监控影像和图像资源的九里湖湿地动植物资源智能识别和监测的系统，并建立了小型的九里湖国家湿地公园生态资源数据集。改进的YOLOV5-X算法，对算法的损失函数和非极大值抑制（Non Maximum Suppression，NMS）算法进行优化，并以接口的方式实现针对特定湿地内的生态环境样本的目标检测功能。对比主流目标识别算法，改进后的YOLOV5-X算法的对动植物识别和视频监测的精度更高，比原始YOLOV5算法识别精度提高了5个百分点，效果更好。YOLOV5-X算法的使用，显著提高了算法目标识别的准确率，能够满足九里湖湿地生态公园智能识别服务的需求。

另外，对湿地保护离不开对湿地环境的监测，考虑到九里湖湿地生态环境复杂，目前对其生态环境监测的方法不够规范，本研究根据研究区湿地生态系统的特点和建设九里湖湿地公园的目标，在系统中加入了一系列能够敏感且清晰反应九里湖湿地生态系统基本特征和九里湖湿地生态环境变化趋势的环境监测指标，并利用长短时人工神经网络（Long short-term memory, LSTM）开展湿地生态环境数据的预测分析。LSTM通过对传统循环神经网络（Recurrent Neural Net, RNN）的改进，使得算法能够有很好的长时间序列的学习能力，对于预测较长时间数据比RNN网络有更好的效果。LSTM的使用，使得平台管理人员可以对监测到的数据进行及时的分析，并对生态环境质量未来的变化趋势进行判断，及时对湿地进行保护。

九里湖湿地生态环境监测系统的建立，为湿地生态保护和修护提供了更加便利的平台，降低了生态环境监测的压力，为湿地生态环境的合理应用、保护和管理提供了依据。

In recent years, with the continuous utilization and transformation of nature by human beings, ecological environment problems are becoming more and more serious. While promoting the sound and rational development of the national economy, China is also paying more attention to the construction of ecological civilization. As an important ecological environment resource, wetland plays an extremely important role in ecological diversity, water conservation, preventing land desertification and reducing greenhouse effect. Therefore, the monitoring and protection of wetland is very important, and the construction of wetland ecological environment monitoring system is one of the important measures to protect wetland resources. The traditional artificial monitoring method has high cost and low efficiency, which can no longer meet the needs of modern platform, information and intelligent wetland ecological environment monitoring. With the continuous introduction of remote sensing technology, wireless sensor technology, artificial intelligence and other technologies into wetland ecological environment monitoring, gradually replace the early artificial monitoring means, improve the efficiency and reliability of wetland ecological environment monitoring system. Based on wireless sensor technology and artificial intelligence technology, this study constructed a wetland ecological environment monitoring system (hereinafter referred to as the system) in Jiulihu National Wetland Park.

With the improvement of the ecological environment of Jiuli Lake, a rich wetland community has been formed, so it is very important to protect the biodiversity in the research area. In this study, an improved target monitoring algorithm (You Only Look Once V5, YOLOV5) was used to develop an intelligent recognition and monitoring system of jiuli Lake wetland animal and plant resources based on monitoring images and image resources, and a small ecological resource data set of Jiuli Lake National Wetland Park was established. The improved YOLOV5-X algorithm optimized the loss function and non-maximum Suppression (NMS) algorithm of the algorithm, and realized the target detection function of ecological environment samples in specific wetlands by interface. Compared with the mainstream target recognition algorithm, the improved YOLOV5-X algorithm has a higher accuracy of animal and plant recognition and video monitoring, which is 5 percentage points higher than the original YOLOV5 algorithm. The use of yOLOV5-X algorithm significantly improves the accuracy of algorithm target recognition, which can meet the needs of jiuli Lake Wetland Ecological Park intelligent identification service.

In addition, the protection of wetland is inseparable from the monitoring of wetland environment. Considering the complex ecological environment of Jiuli Lake wetland, the current monitoring method of its ecological environment is not standardized enough. According to the characteristics of the wetland ecosystem in the research area and the goal of building Jiuli Lake Wetland Park, A series of environmental monitoring indexes were added to the system, which could clearly and sensitively reflect the basic characteristics of jiuli Lake wetland ecosystem and the change trend of jiuli Lake wetland ecological environment, and Long short-term memory (LSTM) was used to predict and analyze the wetland ecological environment data. LSTM has a Recurrent Neural Net (RNN) improvement, which enables the algorithm to have a good learning ability of long time series, and it has a better effect in predicting long time data than RNN network. The use of LSTM enables platform managers to analyze the monitored data in time, judge the future change trend of ecological environment quality, and protect the wetland in time.

The establishment of jiuli Lake wetland ecological environment monitoring system provides a more convenient platform for wetland ecological protection and repair, reduces the pressure of ecological environment monitoring, and provides a basis for the rational application, protection and management of wetland ecological environment.