mkechera/Aquatic-weeds-detection-using-deep-learning-

We have noticed a huge increase of aquatic weeds to different water sources around the world, for example Lake Tanganyika and Lake Victoria in eastern part of Africa. The growth of weeds to the different water sources causes many negative impacts like environment impact, navigation and transport impact, economic impact and agricultural impact. Most of the countries around the world have been trying much to reduce or to eliminate the growth of aquatic weeds to the water bodies, but it happened to be a very great challenge due to the fastest growth of aquatic weeds. Some methods have been applied to eliminate weeds from water sources, for example to transplant insects to the affected areas to consume weeds or by peeling, drying and burning weeds which is very tedious job and time consuming but all these tradition methods have been facing the same challenge on locating and differentiating aquatic weeds with other aquatic plants which are sometimes useful for aquatic life. On this study we have identify the suitable and the most efficiency CNN model that can recognize and allocate aquatic weeds within a short period of time. We have identified YOLOv3 and YOLOv4 as the most suitable algorithms for detecting aquatic weeds and we managed to conduct several scientific experiments basing on the model accuracy of YOLOv3 and YOLOv4 and the result that we found out were 87.52% for YOLOv3 and 93.07% for YOLOv4 that gives us the conclusion that the YOLOv4 is the best deep learning model for detection of the aquatic weeds .But we have seen that the 93% percent accuracy of our model was not enough for the accurate detection of the aquatic weeds ,therefore this paper proposed a method that will increase the accuracy of our model, with the proposed method we have performed some experiments again and the accuracy has been increased by 4.62% that resulted to 97.69% accuracy of our model. # Aquatic-weeds-detection-using-deep-learning- We have noticed a huge increase of aquatic weeds to different water sources around the world, for example Lake Tanganyika and Lake Victoria in eastern part of Africa. The growth of weeds to the different water sources causes many negative impacts like environment impact, navigation and transport impact, economic impact and agricultural impact. Most of the countries around the world have been trying much to reduce or to eliminate the growth of aquatic weeds to the water bodies, but it happened to be a very great challenge due to the fastest growth of aquatic weeds. Some methods have been applied to eliminate weeds from water sources, for example to transplant insects to the affected areas to consume weeds or by peeling, drying and burning weeds which is very tedious job and time consuming but all these tradition methods have been facing the same challenge on locating and differentiating aquatic weeds with other aquatic plants which are sometimes useful for aquatic life. On this study we have identified the suitable and the most efficiency deep learning model that can recognize and allocate aquatic weeds within a short period of time. We have identified YOLOv3 and YOLOv4 as the most suitable algorithms for detecting aquatic weeds and we managed to conduct several scientific experiments basing on the model accuracy of YOLOv3 and YOLOv4 and the result that we found out were 87.52% for YOLOv3 and 93.07% for YOLOv4 that gives us the conclusion that the YOLOv4 is the best deep learning model for detection of the aquatic weeds .But we have seen that the 93% percent accuracy of our model was not enough for the accurate detection of the aquatic weeds ,therefore this paper proposed a method and a deep learning algorithm, that will increase the accuracy of our model, with the proposed methods we have performed some experiments again and the accuracy has been increased by 4.62% that resulted to 97.69% accuracy of our model.