This study presents a novel method for the early detection of Phelipanche ramosa, a serious threat to the California tomato industry, which supplies more than 90% of US processed tomatoes. Due to the underground nature of Phelipanche ramosa, conventional chemical control methods are costly, environmentally harmful, and ineffective. Therefore, this study proposes a method for early detection of Phelipanche ramosa by combining drone-based multispectral imagery with a long short-term memory (LSTM) deep learning network. Specifically, the Synthetic Minority Oversampling Technique (SMOTE) was used to address class imbalance. The study was conducted at five key growth stages in a Phelipanche ramosa-infected tomato farm in Woodland, Yolo County, California, and tomato canopy reflectance was isolated using multispectral image processing. We were able to detect branch foot gall disease with an overall accuracy of 79.09% and recall of 70.36% over 897 GDD (Growing Degree Days). Integrating sequential growth stages using LSTM significantly improved detection performance. The optimal scenario, which integrated all growth stages and SMOTE augmentation, achieved an overall accuracy of 88.37% and recall of 95.37%. This demonstrates the effectiveness of temporal multispectral analysis and LSTM networks in early detection of branch foot gall disease. While additional data collection is required for practical field applications, this study suggests that the combination of UAV-based multispectral sensing and deep learning can be a powerful precision agriculture tool to reduce losses and enhance sustainability in tomato production.
