農地違章工廠對社會造成多方面的影響，尤其是對農業的影響，削弱了農業生產的能力。同時，2022年的俄烏戰爭導致國際糧食價格上漲，台灣作為糧食進口國受到影響，畜牧業的飼料成本增加，進而影響國內畜產品價格。農業不僅提供糧食，還為其他產業提供基本原物料，在經濟發展中扮演著重要角色。為確保國內糧食供應穩定，國家應控制農業產量，避免因國際市場波動或自然災害等因素而導致糧食供應問題。然而，目前國內可供糧食生產的土地面積尚未達到需求，且農地違章工廠佔據農業用地並造成環境汙染，危害國民健康。 在目前深度學習技術廣泛使用，在各個領域都可以看見應用的場合，應用深度學習技術能夠減少大量重複性工作，加速清查作業的進行，因此本研究利用街景影像進行物件偵測，找出建在農地上的違章工廠。本研究以農業及農地資源盤整結果以及google街景圖建立資料集，訓練Mask R-CNN神經網路模型，應用物件偵測於農地違章工廠，以混淆矩陣計算評估模型，F1-score為0.483，預測遮罩平均IoU為0.391。 Illegal factories on farmland have multiple impacts on society, especially on agriculture, as they weaken agricultural productivity. Additionally, the 2022 Russia-Ukraine war led to a surge in international food prices, affecting Taiwan as a food-importing country and causing increased feed costs for the livestock industry, which in turn impacts domestic livestock product prices. Agriculture not only provides food but also serves as a crucial source of raw materials for other industries, playing a vital role in economic development. To ensure stable domestic food supply and mitigate the risks of international market fluctuations or natural disasters, the government should control agricultural production. However, the current available land for food production falls short of the required quantity, and the presence of unauthorized factories on agricultural land not only occupies valuable agricultural resources but also contributes to environmental pollution, posing risks to public health. With the widespread use of deep learning technology and its application in various fields, it has become possible to reduce repetitive tasks and expedite investigation processes. Therefore, this study employs street view imagery for object detection to identify Illegal factories on farmland. The dataset is established using the results of agricultural and land resource surveys and Google Street View images. A Mask R-CNN neural network model is trained for object detection on Illegal factories on farmland. The model's performance is evaluated using a confusion matrix, resulting in an mAP of 0.483 and an average IoU of 0.391 for predicted masks.
