導入自動化與人工智慧技術應用於商用瓢蟲生產 -以瓢蟲成蟲產卵誘導與蟲卵分隔為例

Abstract

瓢蟲（ladybird, ladybug）自 19 世紀末即被廣泛應用於農業害蟲防治，至今已有超過 212 種被引入各類生物防治專案中（Rondoni et al., 2021），為全球重要的生物防治劑（Biological Control Agent, BCA）。隨著有機農業與永續農業理念的推廣，捕食性益蟲於全球生物防治市場預估將維持約 7% 的年成長率（Mordor Intelligence, 2023）。然而，瓢蟲於商業化生產過程中仍面臨高人力成本與生產效率低落等關鍵瓶頸，使其產品價格顯著高於化學農藥，進而降低農民的採用意願（Custom Market Insights, 2023）。針對仍處於市場探索階段、對設備成本具高度敏感性的瓢蟲生物防治量產需求，本研究設計並實作三項可模組化整合之低成本自動化裝置，包含產卵誘導盒（Oviposition Box）、分離與蒐集系統（Isolation and Collection Unit），以及自動輸送模組（Automated Transfer Unit）。 整合後之自動化系統可有效取代傳統高耗時之人工流程，使每 100 顆卵之處理時間顯著縮短，並同時降低卵破損率。各模組均採用 3D 列印結構並結合演算法設計：產卵誘導盒可引導瓢蟲於平滑壓克力基質上集中產卵；分離與蒐集模組則透過紅外線感測器與舵機馬達構成往復運動裝置，利用溫和水流實現卵粒之分離與回收；於卵輸送與辨識模組中，本研究考量早期市場對機械成本之嚴格限制，以成本低於 200 美元之高誤差 robotic arm kit，取代約 30,000 美元之高精度工業用機械手臂（Markvicka et al., 2023），並於結構上結合線性螺桿移動平台與高度限制機制，配合模擬退火（Simulated Annealing）與強化學習（Reinforcement Learning），完成毫米尺度之精準吸卵操作。在目標物辨識策略方面，本研究依據光源穩定性與時間成本需求提出最佳化配置方案：於環境光源穩定條件下，影像處理（Image Processing）方法可達最短計算時間，並結合 GrabCut 影像分割技術以精準描繪卵粒輪廓；於光源不穩定且未使用 GrabCut 之情境中，UNet++ 可提供較細緻之邊緣特徵計算；而在結合 GrabCut 條件下，YOLOv11-seg 則於計算時間與邊緣細緻度之間取得最佳平衡。 為驗證系統於不同飼養行為、生物特性與產卵模式下之適用性，本研究選用目前於台灣以人力方式量產之六條瓢蟲（Cheilomenes sexmaculata），以及具高度捕食潛力之台灣原生種小十三星瓢蟲（Harmonia dimidiata）進行跨模組實驗。結果顯示，小十三星瓢蟲需採獨居飼養以避免同類相食並顯著提升成蟲率（p < 0.001），而六條瓢蟲則可進行群養。在產卵誘導盒中，兩物種於三日內之平均產卵量分別為 25.82 ± 3.89 與 16.67 ± 3.54 顆，目標區域產卵比率分別為 0.64 ± 0.06 與 0.78 ± 0.09。分離與蒐集模組之卵回收率接近 100%，並可透過 pre-immersing 前處理有效降低卵黏連情形，同時維持孵化率。自動輸送模組於精準取卵操作中之成功率超過九成，顯示其具備實際取代人力作業之能力。此外，本研究依據不同物種之生物特性，進一步提出差異化之量產策略。對於攻擊性較低、相食風險小之六條瓢蟲，採群養模式並搭配產卵誘導盒與分離模組，可使每 100 顆卵之處理成本由 24.07 ± 8.33 元新台幣降低至低於 0.01 元；對於需獨居飼養之小十三星瓢蟲，整合三模組後，其處理成本亦可由 57.57 ± 6.81 元降至約 0.02 元。整體自動化流程亦可穩定育成成蟲，每對親蟲於三日內可分別產生 5.19 ± 1.61 與 12.51 ± 2.26 隻成蟲，展現高度穩定性與量產潛力。 綜合而言，本研究所構建之模組化低成本自動化平台，能有效降低人工成本，同時兼具操作精準性、系統穩定性與擴展彈性，不僅適用於瓢蟲之人工繁殖，亦具備推廣至其他小型昆蟲量產系統之潛力，為生物防治技術之規模化、商業化與永續應用提供具體且可行之解決方案。

Ladybirds (ladybird, ladybug) have been widely applied in agricultural pest con-trol since the late nineteenth century, with more than 212 species having been intro-duced into various biological control programs to date (Rondoni et al., 2021), making them globally important Biological Control Agents (BCAs). With the promotion of organic and sustainable agriculture, predatory beneficial insects are expected to main-tain an annual growth rate of approximately 7% in the global biological control market (Mordor Intelligence, 2023). However, the commercial production of ladybirds con-tinues to face critical bottlenecks, including high labor costs and low production effi-ciency, resulting in product prices that are substantially higher than those of chemical pesticides and consequently reducing farmers’ willingness to adopt them (Custom Market Insights, 2023). In response to the mass-production demands of ladybird-based biological control, which remains in an early market exploration stage and is highly sensitive to equipment costs, this study designed and implemented three low-cost, modular, and integrable automated devices, including an Oviposition Box, an Isolation and Collection Unit, and an Automated Transfer Unit. The integrated automated system effectively replaces conventional la-bor-intensive manual procedures, significantly reducing the processing time required per 100 eggs while simultaneously decreasing egg damage rates. All modules adopt 3D-printed structures combined with algorithmic design. The Oviposition Box guides ladybirds to concentrate oviposition on a smooth acrylic substrate. The Isolation and Collection Unit employs an infrared sensor and a servo motor to construct a recipro-cating mechanism that utilizes gentle water flow to achieve egg separation and recov-ery. For egg transfer and recognition, considering the stringent constraints on mechan-ical costs in early-stage markets, this study adopts a high-error robotic arm kit costing less than USD 200 to replace high-precision industrial robotic arms priced at approxi-mately USD 30,000 (Markvicka et al., 2023). By structurally integrating a linear lead-screw motion platform and a height-limiting mechanism, and by incorporating Simulated Annealing and Reinforcement Learning, millimeter-scale precision egg-picking operations are achieved. Regarding target recognition strategies, this study proposes optimized configurations based on lighting stability and time-cost re-quirements. Under stable ambient lighting conditions, image processing methods achieve the shortest computation time and are combined with the GrabCut image seg-mentation technique to accurately delineate egg contours. Under unstable lighting conditions without GrabCut, UNet++ provides more refined edge feature extraction. When combined with GrabCut, YOLOv11-seg achieves the optimal balance between computational time and edge detail. To verify the applicability of the system across different rearing behaviors, bio-logical characteristics, and oviposition patterns, this study selected six-striped lady-birds (Cheilomenes sexmaculata), which are currently mass-produced manually in Taiwan, and the native Taiwanese species small thirteen-spotted ladybird (Harmonia dimidiata), which exhibits high predatory potential, for cross-module experiments. The results indicate that Harmonia dimidiata requires solitary rearing to avoid canni-balism and to significantly improve adult emergence rates (p < 0.001), whereas Cheilomenes sexmaculata can be reared in groups. Within the Oviposition Box, the average numbers of eggs laid by the two species within three days were 25.82 ± 3.89 and 16.67 ± 3.54, respectively, with target-area oviposition ratios of 0.64 ± 0.06 and 0.78 ± 0.09. The egg recovery rate of the Isolation and Collection Unit approached 100%, and pre-immersing pretreatment effectively reduced egg adhesion while main-taining hatchability. The Automated Transfer Unit achieved a success rate exceeding 90% in precision egg-picking operations, demonstrating its practical capability to re-place manual labor. Furthermore, based on the biological characteristics of different species, this study proposes differentiated mass-production strategies. For Cheilomenes sexmaculata, which exhibits lower aggressiveness and a reduced risk of cannibalism, group rearing combined with the Oviposition Box and Isolation and Col-lection Unit reduces the processing cost per 100 eggs from NT$ 24.07 ± 8.33 to below NT$ 0.01. For Harmonia dimidiata, which requires solitary rearing, integration of all three modules reduces the processing cost from NT$ 57.57 ± 6.81 to approximately NT$ 0.02. The overall automated workflow also enables stable adult production, with each parental pair producing 5.19 ± 1.61 and 12.51 ± 2.26 adults within three days, re-spectively, demonstrating high stability and mass-production potential. In summary, the modular low-cost automated platform developed in this study effectively reduces labor costs while maintaining operational precision, system stabil-ity, and scalability. In addition to supporting artificial breeding of ladybirds, the plat-form also shows strong potential for extension to the mass production of other small insects, providing a concrete and feasible solution for the scaling, commercialization, and sustainable application of biological control technologies.