In controlled environment agriculture (CEA), maintaining effective plant spacing throughout the crop growth cycle is crucial for efficient resource (light, water, space, and nutrients) utilization to achieve optimal crop yield and quality. Overcrowded or overlapping plant leaves could cause inefficient light exposure to plants/parts of plants, negatively affecting their growth. Additionally, reduced airflow makes overcrowded plants prone to diseases and foliage damage. Meanwhile, sparse plant spacing could result in inefficient space and light utilization. Traditional plant spacing adjustment relies on expert knowledge and manual labor, which is time-consuming, labor-intensive, and costly. Computer vision-based automatic plant space adjustment could help with data-driven decision-making and reduce labor dependency. This study aims to develop a deep learning-based computer vision approach to estimate the effective plant spacing by extracting the morphological characteristics of plants and NFT (nutrient film technique) channels during different plant growth stages. A total of 576 lettuce plants were grown in an NFT channel-based hydroponics system in a controlled environment. Then, RGB-D information of these plants and NFT channels was collected each day for three weeks from planting to harvesting. Then, CNN (convolutional neural network) was employed to extract the plant and NFT channel feature information. Then, the spatial pyramid pooling approach was used to encode and decode the contextual information and segment the plants and NFT channels. This approach helped to achieve an F1-score of 0.90 on the test dataset to estimate space between plants and NFT channels. These results show the potential of the proposed approach for automated plant space adjustment for efficient resource utilization.
