Forecasting flower varieties and identifying impaired flower fields through image segmentation and neural networks

Floriculture can be vital in boosting farmer's income and the country's economy. In recent decades, the demand for loose and cut flowers has rapidly increased in both national and international markets. Despite having a great scope in this sector, farmers face challenges such as a lack of cold storage, adaptation to advanced technologies, regular maintenance, poor marketing strategies, and lack of entrepreneurship. A significant amount of the cultivated area remains underdeveloped or damaged in flower fields due to a lack of regular supervision of the cultivation area. Although manually monitoring large fields is tedious and inefficient, image segmentation and an artificial neural network can make the process automated and mitigate the challenges. Therefore, this paper presents an approach for the prediction of flower variety and the detection of underdeveloped or impaired flower fields using image segmentation and a neural network model. The proposed approach employs a collection of pixels RGB values extracted from drone images and a list of crop varieties as input and expected output respectively to train and test the neural network. Holistically nested edge detection (HED) and connected component-based labelling (CCL) are utilized to detect underdeveloped or impaired flower field areas. With the significant precision and efficiency of the proposed approach, it provides a solution to the lack of regular supervision of cultivation areas and outperforms recent baseline methods. • Drone-based data collection for flower growth in target regions. • Analyze image data to support flower growth monitoring. • Predict flower types and cluster distribution using neural networks. • Detect impaired flower fields using prediction and edge detection.